Compare commits

..

1 Commits

@ -1,22 +0,0 @@
name: Bindings Tests (Ruby)
on:
push:
paths:
- bindings/ruby/**
- whisper.h
pull_request:
paths:
- bindings/ruby/**
- whisper.h
jobs:
ubuntu-latest:
runs-on: ubuntu-latest
steps:
- uses: ruby/setup-ruby@v1
with:
ruby-version: '3.0'
- uses: actions/checkout@v1
- run: |
cd bindings/ruby/ext
ruby extconf.rb && make

@ -1,18 +1,13 @@
name: Bindings Tests (Go)
name: Bindings Tests
on:
push:
paths:
- bindings/go/**
- whisper.h
pull_request:
paths:
- bindings/go/**
- whisper.h
jobs:
ubuntu-latest:
runs-on: ubuntu-latest
steps:
ubuntu-latest:
runs-on: ubuntu-latest
steps:
- uses: actions/setup-go@v3
with:
go-version: '^1.19'

@ -1,267 +1,267 @@
name: CI
on: [push, pull_request]
on: [push]
jobs:
ubuntu-latest:
runs-on: ubuntu-latest
steps:
- name: Clone
uses: actions/checkout@v1
- name: Dependencies
run: |
sudo apt-get update
sudo apt-get install build-essential
sudo apt-get install libsdl2-dev
- name: Build
run: |
make
make stream
macOS-latest:
runs-on: macOS-latest
steps:
- name: Clone
uses: actions/checkout@v1
- name: Dependencies
run: |
brew update
brew install sdl2
- name: Build
run: |
make
make stream
ubuntu-latest-gcc:
runs-on: ubuntu-latest
strategy:
matrix:
build: [Debug, Release]
steps:
- name: Clone
uses: actions/checkout@v1
- name: Dependencies
run: |
sudo apt-get update
sudo apt-get install build-essential
sudo apt-get install cmake
sudo apt-get install libsdl2-dev
- name: Configure
run: cmake . -DWHISPER_SUPPORT_SDL2=ON -DCMAKE_BUILD_TYPE=${{ matrix.build }}
- name: Build
run: |
make
ctest -L gh --output-on-failure
ubuntu-latest-clang:
runs-on: ubuntu-latest
strategy:
matrix:
build: [Debug, Release]
steps:
- name: Clone
uses: actions/checkout@v1
- name: Dependencies
run: |
sudo apt-get update
sudo apt-get install build-essential
sudo apt-get install cmake
sudo apt-get install libsdl2-dev
- name: Configure
run: cmake . -DWHISPER_SUPPORT_SDL2=ON -DCMAKE_BUILD_TYPE=${{ matrix.build }} -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_C_COMPILER=clang
- name: Build
run: |
make
ctest -L gh --output-on-failure
ubuntu-latest-gcc-sanitized:
runs-on: ubuntu-latest
strategy:
matrix:
sanitizer: [ADDRESS, THREAD, UNDEFINED]
steps:
- name: Clone
uses: actions/checkout@v1
- name: Dependencies
run: |
sudo apt-get update
sudo apt-get install build-essential
sudo apt-get install cmake
- name: Configure
run: cmake . -DCMAKE_BUILD_TYPE=Debug -DWHISPER_SANITIZE_${{ matrix.sanitizer }}=ON
- name: Build
run: |
make
ctest -L gh --output-on-failure
windows:
runs-on: windows-latest
strategy:
matrix:
build: [Release]
arch: [Win32, x64]
sdl2: [ON]
include:
- arch: Win32
s2arc: x86
- arch: x64
s2arc: x64
- sdl2: ON
s2ver: 2.26.0
steps:
- name: Clone
uses: actions/checkout@v1
- name: Add msbuild to PATH
uses: microsoft/setup-msbuild@v1
- name: Fetch SDL2 and set SDL2_DIR
if: matrix.sdl2 == 'ON'
run: |
C:/msys64/usr/bin/wget.exe -qO sdl2.zip https://github.com/libsdl-org/SDL/releases/download/release-${{ matrix.s2ver }}/SDL2-devel-${{ matrix.s2ver }}-VC.zip
7z x sdl2.zip
echo "SDL2_DIR=$env:GITHUB_WORKSPACE/SDL2-${{ matrix.s2ver }}/cmake" >> $env:GITHUB_ENV
- name: Configure
run: >
cmake -S . -B ./build -A ${{ matrix.arch }}
-DCMAKE_BUILD_TYPE=${{ matrix.build }}
-DWHISPER_SUPPORT_SDL2=${{ matrix.sdl2 }}
- name: Build
run: |
cd ./build
msbuild ALL_BUILD.vcxproj -t:build -p:configuration=${{ matrix.build }} -p:platform=${{ matrix.arch }}
- name: Copy SDL2.dll
if: matrix.sdl2 == 'ON'
run: copy "$env:SDL2_DIR/../lib/${{ matrix.s2arc }}/SDL2.dll" build/bin/${{ matrix.build }}
- name: Upload binaries
if: matrix.sdl2 == 'ON'
uses: actions/upload-artifact@v1
with:
name: whisper-bin-${{ matrix.arch }}
path: build/bin/${{ matrix.build }}
windows-blas:
runs-on: windows-latest
strategy:
matrix:
build: [Release]
arch: [Win32, x64]
blas: [ON]
sdl2: [ON]
include:
- arch: Win32
obzip: https://github.com/xianyi/OpenBLAS/releases/download/v0.3.21/OpenBLAS-0.3.21-x86.zip
s2arc: x86
- arch: x64
obzip: https://github.com/xianyi/OpenBLAS/releases/download/v0.3.21/OpenBLAS-0.3.21-x64.zip
s2arc: x64
- sdl2: ON
s2ver: 2.26.0
steps:
- name: Clone
uses: actions/checkout@v1
- name: Add msbuild to PATH
uses: microsoft/setup-msbuild@v1
- name: Fetch OpenBLAS
if: matrix.blas == 'ON'
run: |
C:/msys64/usr/bin/wget.exe -qO blas.zip ${{ matrix.obzip }}
7z x blas.zip -oblas -y
copy blas/include/cblas.h .
copy blas/include/openblas_config.h .
echo "blasdir=$env:GITHUB_WORKSPACE/blas" >> $env:GITHUB_ENV
- name: Fetch SDL2 and set SDL2_DIR
if: matrix.sdl2 == 'ON'
run: |
C:/msys64/usr/bin/wget.exe -qO sdl2.zip https://github.com/libsdl-org/SDL/releases/download/release-${{ matrix.s2ver }}/SDL2-devel-${{ matrix.s2ver }}-VC.zip
7z x sdl2.zip
echo "SDL2_DIR=$env:GITHUB_WORKSPACE/SDL2-${{ matrix.s2ver }}/cmake" >> $env:GITHUB_ENV
- name: Configure
run: >
cmake -S . -B ./build -A ${{ matrix.arch }}
-DCMAKE_BUILD_TYPE=${{ matrix.build }}
-DWHISPER_SUPPORT_OPENBLAS=${{ matrix.blas }}
-DCMAKE_LIBRARY_PATH="$env:blasdir/lib"
-DWHISPER_SUPPORT_SDL2=${{ matrix.sdl2 }}
- name: Build
run: |
cd ./build
msbuild ALL_BUILD.vcxproj -t:build -p:configuration=${{ matrix.build }} -p:platform=${{ matrix.arch }}
- name: Copy libopenblas.dll
if: matrix.blas == 'ON'
run: copy "$env:blasdir/bin/libopenblas.dll" build/bin/${{ matrix.build }}
- name: Copy SDL2.dll
if: matrix.sdl2 == 'ON'
run: copy "$env:SDL2_DIR/../lib/${{ matrix.s2arc }}/SDL2.dll" build/bin/${{ matrix.build }}
- name: Upload binaries
if: matrix.blas == 'ON' && matrix.sdl2 == 'ON'
uses: actions/upload-artifact@v1
with:
name: whisper-blas-bin-${{ matrix.arch }}
path: build/bin/${{ matrix.build }}
emscripten:
runs-on: ubuntu-latest
strategy:
matrix:
build: [Release]
steps:
- name: Clone
uses: actions/checkout@v1
- name: Dependencies
run: |
wget -q https://github.com/emscripten-core/emsdk/archive/master.tar.gz
tar -xvf master.tar.gz
emsdk-master/emsdk update
emsdk-master/emsdk install latest
emsdk-master/emsdk activate latest
- name: Configure
run: echo "tmp"
- name: Build
run: |
pushd emsdk-master
source ./emsdk_env.sh
popd
emcmake cmake . -DCMAKE_BUILD_TYPE=${{ matrix.build }}
make
ubuntu-latest:
runs-on: ubuntu-latest
steps:
- name: Clone
uses: actions/checkout@v1
- name: Dependencies
run: |
sudo apt-get update
sudo apt-get install build-essential
sudo apt-get install libsdl2-dev
- name: Build
run: |
make
make stream
macOS-latest:
runs-on: macOS-latest
steps:
- name: Clone
uses: actions/checkout@v1
- name: Dependencies
run: |
brew update
brew install sdl2
- name: Build
run: |
make
make stream
ubuntu-latest-gcc:
runs-on: ubuntu-latest
strategy:
matrix:
build: [Debug, Release]
steps:
- name: Clone
uses: actions/checkout@v1
- name: Dependencies
run: |
sudo apt-get update
sudo apt-get install build-essential
sudo apt-get install cmake
sudo apt-get install libsdl2-dev
- name: Configure
run: cmake . -DWHISPER_SUPPORT_SDL2=ON -DCMAKE_BUILD_TYPE=${{ matrix.build }}
- name: Build
run: |
make
ctest -L gh --output-on-failure
ubuntu-latest-clang:
runs-on: ubuntu-latest
strategy:
matrix:
build: [Debug, Release]
steps:
- name: Clone
uses: actions/checkout@v1
- name: Dependencies
run: |
sudo apt-get update
sudo apt-get install build-essential
sudo apt-get install cmake
sudo apt-get install libsdl2-dev
- name: Configure
run: cmake . -DWHISPER_SUPPORT_SDL2=ON -DCMAKE_BUILD_TYPE=${{ matrix.build }} -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_C_COMPILER=clang
- name: Build
run: |
make
ctest -L gh --output-on-failure
ubuntu-latest-gcc-sanitized:
runs-on: ubuntu-latest
strategy:
matrix:
sanitizer: [ADDRESS, THREAD, UNDEFINED]
steps:
- name: Clone
uses: actions/checkout@v1
- name: Dependencies
run: |
sudo apt-get update
sudo apt-get install build-essential
sudo apt-get install cmake
- name: Configure
run: cmake . -DCMAKE_BUILD_TYPE=Debug -DWHISPER_SANITIZE_${{ matrix.sanitizer }}=ON
- name: Build
run: |
make
ctest -L gh --output-on-failure
windows:
runs-on: windows-latest
strategy:
matrix:
build: [Release]
arch: [Win32, x64]
sdl2: [ON]
include:
- arch: Win32
s2arc: x86
- arch: x64
s2arc: x64
- sdl2: ON
s2ver: 2.26.0
steps:
- name: Clone
uses: actions/checkout@v1
- name: Add msbuild to PATH
uses: microsoft/setup-msbuild@v1
- name: Fetch SDL2 and set SDL2_DIR
if: matrix.sdl2 == 'ON'
run: |
C:/msys64/usr/bin/wget.exe -qO sdl2.zip https://github.com/libsdl-org/SDL/releases/download/release-${{ matrix.s2ver }}/SDL2-devel-${{ matrix.s2ver }}-VC.zip
7z x sdl2.zip
echo "SDL2_DIR=$env:GITHUB_WORKSPACE/SDL2-${{ matrix.s2ver }}/cmake" >> $env:GITHUB_ENV
- name: Configure
run: >
cmake -S . -B ./build -A ${{ matrix.arch }}
-DCMAKE_BUILD_TYPE=${{ matrix.build }}
-DWHISPER_SUPPORT_SDL2=${{ matrix.sdl2 }}
- name: Build
run: |
cd ./build
msbuild ALL_BUILD.vcxproj -t:build -p:configuration=${{ matrix.build }} -p:platform=${{ matrix.arch }}
- name: Copy SDL2.dll
if: matrix.sdl2 == 'ON'
run: copy "$env:SDL2_DIR/../lib/${{ matrix.s2arc }}/SDL2.dll" build/bin/${{ matrix.build }}
- name: Upload binaries
if: matrix.sdl2 == 'ON'
uses: actions/upload-artifact@v1
with:
name: whisper-bin-${{ matrix.arch }}
path: build/bin/${{ matrix.build }}
windows-blas:
runs-on: windows-latest
strategy:
matrix:
build: [Release]
arch: [Win32, x64]
blas: [ON]
sdl2: [ON]
include:
- arch: Win32
obzip: https://github.com/xianyi/OpenBLAS/releases/download/v0.3.21/OpenBLAS-0.3.21-x86.zip
s2arc: x86
- arch: x64
obzip: https://github.com/xianyi/OpenBLAS/releases/download/v0.3.21/OpenBLAS-0.3.21-x64.zip
s2arc: x64
- sdl2: ON
s2ver: 2.26.0
steps:
- name: Clone
uses: actions/checkout@v1
- name: Add msbuild to PATH
uses: microsoft/setup-msbuild@v1
- name: Fetch OpenBLAS
if: matrix.blas == 'ON'
run: |
C:/msys64/usr/bin/wget.exe -qO blas.zip ${{ matrix.obzip }}
7z x blas.zip -oblas -y
copy blas/include/cblas.h .
copy blas/include/openblas_config.h .
echo "blasdir=$env:GITHUB_WORKSPACE/blas" >> $env:GITHUB_ENV
- name: Fetch SDL2 and set SDL2_DIR
if: matrix.sdl2 == 'ON'
run: |
C:/msys64/usr/bin/wget.exe -qO sdl2.zip https://github.com/libsdl-org/SDL/releases/download/release-${{ matrix.s2ver }}/SDL2-devel-${{ matrix.s2ver }}-VC.zip
7z x sdl2.zip
echo "SDL2_DIR=$env:GITHUB_WORKSPACE/SDL2-${{ matrix.s2ver }}/cmake" >> $env:GITHUB_ENV
- name: Configure
run: >
cmake -S . -B ./build -A ${{ matrix.arch }}
-DCMAKE_BUILD_TYPE=${{ matrix.build }}
-DWHISPER_SUPPORT_OPENBLAS=${{ matrix.blas }}
-DCMAKE_LIBRARY_PATH="$env:blasdir/lib"
-DWHISPER_SUPPORT_SDL2=${{ matrix.sdl2 }}
- name: Build
run: |
cd ./build
msbuild ALL_BUILD.vcxproj -t:build -p:configuration=${{ matrix.build }} -p:platform=${{ matrix.arch }}
- name: Copy libopenblas.dll
if: matrix.blas == 'ON'
run: copy "$env:blasdir/bin/libopenblas.dll" build/bin/${{ matrix.build }}
- name: Copy SDL2.dll
if: matrix.sdl2 == 'ON'
run: copy "$env:SDL2_DIR/../lib/${{ matrix.s2arc }}/SDL2.dll" build/bin/${{ matrix.build }}
- name: Upload binaries
if: matrix.blas == 'ON' && matrix.sdl2 == 'ON'
uses: actions/upload-artifact@v1
with:
name: whisper-blas-bin-${{ matrix.arch }}
path: build/bin/${{ matrix.build }}
emscripten:
runs-on: ubuntu-latest
strategy:
matrix:
build: [Release]
steps:
- name: Clone
uses: actions/checkout@v1
- name: Dependencies
run: |
wget -q https://github.com/emscripten-core/emsdk/archive/master.tar.gz
tar -xvf master.tar.gz
emsdk-master/emsdk update
emsdk-master/emsdk install latest
emsdk-master/emsdk activate latest
- name: Configure
run: echo "tmp"
- name: Build
run: |
pushd emsdk-master
source ./emsdk_env.sh
popd
emcmake cmake . -DCMAKE_BUILD_TYPE=${{ matrix.build }}
make

@ -1,48 +0,0 @@
name: Examples Tests
on:
push:
paths:
- examples/addon.node/**
- whisper.h
pull_request:
paths:
- examples/addon.node/**
- whisper.h
jobs:
addon_node-ubuntu-latest:
runs-on: ubuntu-latest
strategy:
matrix:
node-version: [ 16.x, 18.x ]
steps:
- name: Clone
uses: actions/checkout@v1
- name: Dependencies
run: |
sudo apt-get update
sudo apt-get install build-essential
sudo apt-get install cmake
sudo apt-get install libsdl2-dev
- name: Use Node.js ${{ matrix.node-version }}
uses: actions/setup-node@v1
with:
node-version: ${{ matrix.node-version }}
cache: 'npm'
- name: Install package.json dependencies
working-directory: ./examples/addon.node
run: npm install
- name: Compile addon.node
run: npx cmake-js compile -T whisper-addon -B Release
- name: Download test model
run: |
bash ./models/download-ggml-model.sh base.en
- name: Test
run: |
cd examples/addon.node
npm run test

3
.gitignore vendored

@ -1,5 +1,4 @@
*.o
*.a
.cache/
.vs/
.vscode/
@ -10,7 +9,6 @@ build-em/
build-debug/
build-release/
build-static/
build-no-accel/
build-sanitize-addr/
build-sanitize-thread/
@ -20,7 +18,6 @@ build-sanitize-thread/
/talk
/bench
arm_neon.h
sync.sh
libwhisper.a
libwhisper.so

@ -1,6 +1,6 @@
cmake_minimum_required (VERSION 3.0)
project(whisper.cpp VERSION 1.2.1)
project(whisper.cpp VERSION 1.1.0)
# Add path to modules
list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake/")
@ -226,13 +226,10 @@ target_compile_definitions(${TARGET} PUBLIC
${WHISPER_EXTRA_FLAGS}
)
set_target_properties(${TARGET} PROPERTIES PUBLIC_HEADER "whisper.h")
install(TARGETS ${TARGET}
LIBRARY DESTINATION lib
ARCHIVE DESTINATION lib/static
RUNTIME DESTINATION bin
PUBLIC_HEADER DESTINATION include
)
#
@ -245,7 +242,7 @@ add_subdirectory(bindings)
# programs, examples and tests
#
if (WHISPER_BUILD_TESTS AND NOT CMAKE_JS_VERSION)
if (WHISPER_BUILD_TESTS)
enable_testing()
add_subdirectory(tests)
endif ()

@ -30,8 +30,8 @@ endif
# Compile flags
#
CFLAGS = -I. -O3 -DNDEBUG -std=c11 -fPIC
CXXFLAGS = -I. -I./examples -O3 -DNDEBUG -std=c++11 -fPIC
CFLAGS = -I. -O3 -std=c11 -fPIC
CXXFLAGS = -I. -I./examples -O3 -std=c++11 -fPIC
LDFLAGS =
# OS specific
@ -115,15 +115,11 @@ endif
ifeq ($(UNAME_M),amd64)
CFLAGS += -mavx -mavx2 -mfma -mf16c
endif
ifneq ($(filter ppc64%,$(UNAME_M)),)
ifeq ($(UNAME_M),ppc64le)
POWER9_M := $(shell grep "POWER9" /proc/cpuinfo)
ifneq (,$(findstring POWER9,$(POWER9_M)))
CFLAGS += -mpower9-vector
endif
# Require c++23's std::byteswap for big-endian support.
ifeq ($(UNAME_M),ppc64)
CXXFLAGS += -std=c++23 -DGGML_BIG_ENDIAN
endif
endif
ifndef WHISPER_NO_ACCELERATE
# Mac M1 - include Accelerate framework
@ -141,8 +137,6 @@ ifdef WHISPER_GPROF
CXXFLAGS += -pg
endif
ifneq ($(filter aarch64%,$(UNAME_M)),)
CFLAGS += -mcpu=native
CXXFLAGS += -mcpu=native
endif
ifneq ($(filter armv6%,$(UNAME_M)),)
# Raspberry Pi 1, 2, 3
@ -199,21 +193,18 @@ clean:
CC_SDL=`sdl2-config --cflags --libs`
SRC_COMMON = examples/common.cpp
SRC_COMMON_SDL = examples/common-sdl.cpp
main: examples/main/main.cpp $(SRC_COMMON) ggml.o whisper.o
$(CXX) $(CXXFLAGS) examples/main/main.cpp $(SRC_COMMON) ggml.o whisper.o -o main $(LDFLAGS)
main: examples/main/main.cpp ggml.o whisper.o
$(CXX) $(CXXFLAGS) examples/main/main.cpp ggml.o whisper.o -o main $(LDFLAGS)
./main -h
stream: examples/stream/stream.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) ggml.o whisper.o
$(CXX) $(CXXFLAGS) examples/stream/stream.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) ggml.o whisper.o -o stream $(CC_SDL) $(LDFLAGS)
stream: examples/stream/stream.cpp ggml.o whisper.o
$(CXX) $(CXXFLAGS) examples/stream/stream.cpp ggml.o whisper.o -o stream $(CC_SDL) $(LDFLAGS)
command: examples/command/command.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) ggml.o whisper.o
$(CXX) $(CXXFLAGS) examples/command/command.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) ggml.o whisper.o -o command $(CC_SDL) $(LDFLAGS)
command: examples/command/command.cpp ggml.o whisper.o
$(CXX) $(CXXFLAGS) examples/command/command.cpp ggml.o whisper.o -o command $(CC_SDL) $(LDFLAGS)
talk: examples/talk/talk.cpp examples/talk/gpt-2.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) ggml.o whisper.o
$(CXX) $(CXXFLAGS) examples/talk/talk.cpp examples/talk/gpt-2.cpp $(SRC_COMMON) $(SRC_COMMON_SDL) ggml.o whisper.o -o talk $(CC_SDL) $(LDFLAGS)
talk: examples/talk/talk.cpp examples/talk/gpt-2.cpp ggml.o whisper.o
$(CXX) $(CXXFLAGS) examples/talk/talk.cpp examples/talk/gpt-2.cpp ggml.o whisper.o -o talk $(CC_SDL) $(LDFLAGS)
bench: examples/bench/bench.cpp ggml.o whisper.o
$(CXX) $(CXXFLAGS) examples/bench/bench.cpp ggml.o whisper.o -o bench $(LDFLAGS)

@ -4,7 +4,7 @@
[![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](https://opensource.org/licenses/MIT)
[![npm](https://img.shields.io/npm/v/whisper.cpp.svg)](https://www.npmjs.com/package/whisper.cpp/)
Stable: [v1.2.1](https://github.com/ggerganov/whisper.cpp/releases/tag/v1.2.1) / [Roadmap | F.A.Q.](https://github.com/ggerganov/whisper.cpp/discussions/126)
Stable: [v1.0.4](https://github.com/ggerganov/whisper.cpp/releases/tag/v1.0.4) / Beta: [v1.1.0](https://github.com/ggerganov/whisper.cpp/releases/tag/v1.1.0) / [Roadmap | F.A.Q.](https://github.com/ggerganov/whisper.cpp/discussions/126)
High-performance inference of [OpenAI's Whisper](https://github.com/openai/whisper) automatic speech recognition (ASR) model:
@ -13,7 +13,7 @@ High-performance inference of [OpenAI's Whisper](https://github.com/openai/whisp
- AVX intrinsics support for x86 architectures
- VSX intrinsics support for POWER architectures
- Mixed F16 / F32 precision
- Low memory usage (Flash Attention)
- Low memory usage (Flash Attention + Flash Forward)
- Zero memory allocations at runtime
- Runs on the CPU
- [C-style API](https://github.com/ggerganov/whisper.cpp/blob/master/whisper.h)
@ -89,37 +89,35 @@ c++ -I. -I./examples -O3 -std=c++11 -pthread examples/main/main.cpp whisper.o gg
usage: ./main [options] file0.wav file1.wav ...
options:
-h, --help [default] show this help message and exit
-t N, --threads N [4 ] number of threads to use during computation
-p N, --processors N [1 ] number of processors to use during computation
-ot N, --offset-t N [0 ] time offset in milliseconds
-on N, --offset-n N [0 ] segment index offset
-d N, --duration N [0 ] duration of audio to process in milliseconds
-mc N, --max-context N [-1 ] maximum number of text context tokens to store
-ml N, --max-len N [0 ] maximum segment length in characters
-bo N, --best-of N [5 ] number of best candidates to keep
-bs N, --beam-size N [-1 ] beam size for beam search
-wt N, --word-thold N [0.01 ] word timestamp probability threshold
-et N, --entropy-thold N [2.40 ] entropy threshold for decoder fail
-lpt N, --logprob-thold N [-1.00 ] log probability threshold for decoder fail
-su, --speed-up [false ] speed up audio by x2 (reduced accuracy)
-tr, --translate [false ] translate from source language to english
-di, --diarize [false ] stereo audio diarization
-nf, --no-fallback [false ] do not use temperature fallback while decoding
-otxt, --output-txt [false ] output result in a text file
-ovtt, --output-vtt [false ] output result in a vtt file
-osrt, --output-srt [false ] output result in a srt file
-owts, --output-words [false ] output script for generating karaoke video
-ocsv, --output-csv [false ] output result in a CSV file
-of FNAME, --output-file FNAME [ ] output file path (without file extension)
-ps, --print-special [false ] print special tokens
-pc, --print-colors [false ] print colors
-pp, --print-progress [false ] print progress
-nt, --no-timestamps [true ] do not print timestamps
-l LANG, --language LANG [en ] spoken language ('auto' for auto-detect)
--prompt PROMPT [ ] initial prompt
-m FNAME, --model FNAME [models/ggml-base.en.bin] model path
-f FNAME, --file FNAME [ ] input WAV file path
-h, --help [default] show this help message and exit
-t N, --threads N [4 ] number of threads to use during computation
-p N, --processors N [1 ] number of processors to use during computation
-ot N, --offset-t N [0 ] time offset in milliseconds
-on N, --offset-n N [0 ] segment index offset
-d N, --duration N [0 ] duration of audio to process in milliseconds
-mc N, --max-context N [-1 ] maximum number of text context tokens to store
-ml N, --max-len N [0 ] maximum segment length in characters
-bo N, --best-of N [5 ] number of best candidates to keep
-bs N, --beam-size N [-1 ] beam size for beam search
-wt N, --word-thold N [0.01 ] word timestamp probability threshold
-et N, --entropy-thold N [2.40 ] entropy threshold for decoder fail
-lpt N, --logprob-thold N [-1.00 ] log probability threshold for decoder fail
-su, --speed-up [false ] speed up audio by x2 (reduced accuracy)
-tr, --translate [false ] translate from source language to english
-di, --diarize [false ] stereo audio diarization
-otxt, --output-txt [false ] output result in a text file
-ovtt, --output-vtt [false ] output result in a vtt file
-osrt, --output-srt [false ] output result in a srt file
-owts, --output-words [false ] output script for generating karaoke video
-ocsv, --output-csv [false ] output result in a CSV file
-ps, --print-special [false ] print special tokens
-pc, --print-colors [false ] print colors
-pp, --print-progress [false ] print progress
-nt, --no-timestamps [true ] do not print timestamps
-l LANG, --language LANG [en ] spoken language ('auto' for auto-detect)
--prompt PROMPT [ ] initial prompt
-m FNAME, --model FNAME [models/ggml-base.en.bin] model path
-f FNAME, --file FNAME [ ] input WAV file path
bash ./models/download-ggml-model.sh base.en
@ -139,8 +137,7 @@ Running base.en on all samples in ./samples ...
[+] Running base.en on samples/jfk.wav ... (run 'ffplay samples/jfk.wav' to listen)
----------------------------------------------
whisper_init_from_file: loading model from 'models/ggml-base.en.bin'
whisper_model_load: loading model
whisper_model_load: loading model from 'models/ggml-base.en.bin'
whisper_model_load: n_vocab = 51864
whisper_model_load: n_audio_ctx = 1500
whisper_model_load: n_audio_state = 512
@ -153,14 +150,13 @@ whisper_model_load: n_text_layer = 6
whisper_model_load: n_mels = 80
whisper_model_load: f16 = 1
whisper_model_load: type = 2
whisper_model_load: mem required = 215.00 MB (+ 6.00 MB per decoder)
whisper_model_load: kv self size = 5.25 MB
whisper_model_load: kv cross size = 17.58 MB
whisper_model_load: adding 1607 extra tokens
whisper_model_load: model ctx = 140.60 MB
whisper_model_load: mem_required = 506.00 MB
whisper_model_load: ggml ctx size = 140.60 MB
whisper_model_load: memory size = 22.83 MB
whisper_model_load: model size = 140.54 MB
system_info: n_threads = 4 / 10 | AVX = 0 | AVX2 = 0 | AVX512 = 0 | FMA = 0 | NEON = 1 | ARM_FMA = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 0 | VSX = 0 |
system_info: n_threads = 4 / 10 | AVX = 0 | AVX2 = 0 | AVX512 = 0 | NEON = 1 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 |
main: processing 'samples/jfk.wav' (176000 samples, 11.0 sec), 4 threads, 1 processors, lang = en, task = transcribe, timestamps = 1 ...
@ -168,13 +164,12 @@ main: processing 'samples/jfk.wav' (176000 samples, 11.0 sec), 4 threads, 1 proc
[00:00:00.000 --> 00:00:11.000] And so my fellow Americans, ask not what your country can do for you, ask what you can do for your country.
whisper_print_timings: fallbacks = 0 p / 0 h
whisper_print_timings: load time = 113.81 ms
whisper_print_timings: mel time = 15.40 ms
whisper_print_timings: sample time = 11.58 ms / 27 runs ( 0.43 ms per run)
whisper_print_timings: encode time = 266.60 ms / 1 runs ( 266.60 ms per run)
whisper_print_timings: decode time = 66.11 ms / 27 runs ( 2.45 ms per run)
whisper_print_timings: total time = 476.31 ms
whisper_print_timings: load time = 105.91 ms
whisper_print_timings: mel time = 24.62 ms
whisper_print_timings: sample time = 3.63 ms
whisper_print_timings: encode time = 324.71 ms / 54.12 ms per layer
whisper_print_timings: decode time = 83.58 ms / 13.93 ms per layer
whisper_print_timings: total time = 542.81 ms
```
The command downloads the `base.en` model converted to custom `ggml` format and runs the inference on all `.wav` samples in the folder `samples`.
@ -217,11 +212,11 @@ make large
| Model | Disk | Mem | SHA |
| --- | --- | --- | --- |
| tiny | 75 MB | ~125 MB | `bd577a113a864445d4c299885e0cb97d4ba92b5f` |
| base | 142 MB | ~210 MB | `465707469ff3a37a2b9b8d8f89f2f99de7299dac` |
| small | 466 MB | ~600 MB | `55356645c2b361a969dfd0ef2c5a50d530afd8d5` |
| medium | 1.5 GB | ~1.7 GB | `fd9727b6e1217c2f614f9b698455c4ffd82463b4` |
| large | 2.9 GB | ~3.3 GB | `0f4c8e34f21cf1a914c59d8b3ce882345ad349d6` |
| tiny | 75 MB | ~390 MB | `bd577a113a864445d4c299885e0cb97d4ba92b5f` |
| base | 142 MB | ~500 MB | `465707469ff3a37a2b9b8d8f89f2f99de7299dac` |
| small | 466 MB | ~1.0 GB | `55356645c2b361a969dfd0ef2c5a50d530afd8d5` |
| medium | 1.5 GB | ~2.6 GB | `fd9727b6e1217c2f614f9b698455c4ffd82463b4` |
| large | 2.9 GB | ~4.7 GB | `0f4c8e34f21cf1a914c59d8b3ce882345ad349d6` |
## Limitations
@ -239,8 +234,7 @@ in about half a minute on a MacBook M1 Pro, using `medium.en` model:
```java
$ ./main -m models/ggml-medium.en.bin -f samples/gb1.wav -t 8
whisper_init_from_file: loading model from 'models/ggml-medium.en.bin'
whisper_model_load: loading model
whisper_model_load: loading model from 'models/ggml-medium.en.bin'
whisper_model_load: n_vocab = 51864
whisper_model_load: n_audio_ctx = 1500
whisper_model_load: n_audio_state = 1024
@ -253,60 +247,55 @@ whisper_model_load: n_text_layer = 24
whisper_model_load: n_mels = 80
whisper_model_load: f16 = 1
whisper_model_load: type = 4
whisper_model_load: mem required = 1720.00 MB (+ 43.00 MB per decoder)
whisper_model_load: kv self size = 42.00 MB
whisper_model_load: kv cross size = 140.62 MB
whisper_model_load: mem_required = 2610.00 MB
whisper_model_load: adding 1607 extra tokens
whisper_model_load: model ctx = 1462.35 MB
whisper_model_load: model size = 1462.12 MB
system_info: n_threads = 8 / 10 | AVX = 0 | AVX2 = 0 | AVX512 = 0 | FMA = 0 | NEON = 1 | ARM_FMA = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 0 | VSX = 0 |
main: processing 'samples/gb1.wav' (3179750 samples, 198.7 sec), 8 threads, 1 processors, lang = en, task = transcribe, timestamps = 1 ...
[00:00:00.000 --> 00:00:08.000] My fellow Americans, this day has brought terrible news and great sadness to our country.
[00:00:08.000 --> 00:00:17.000] At nine o'clock this morning, Mission Control in Houston lost contact with our Space Shuttle Columbia.
[00:00:17.000 --> 00:00:23.000] A short time later, debris was seen falling from the skies above Texas.
[00:00:23.000 --> 00:00:29.000] The Columbia's lost. There are no survivors.
[00:00:29.000 --> 00:00:32.000] On board was a crew of seven.
[00:00:32.000 --> 00:00:39.000] Colonel Rick Husband, Lieutenant Colonel Michael Anderson, Commander Laurel Clark,
[00:00:39.000 --> 00:00:48.000] Captain David Brown, Commander William McCool, Dr. Kultna Shavla, and Ilan Ramon,
[00:00:48.000 --> 00:00:52.000] a colonel in the Israeli Air Force.
[00:00:52.000 --> 00:00:58.000] These men and women assumed great risk in the service to all humanity.
[00:00:58.000 --> 00:01:03.000] In an age when space flight has come to seem almost routine,
[00:01:03.000 --> 00:01:07.000] it is easy to overlook the dangers of travel by rocket
[00:01:07.000 --> 00:01:12.000] and the difficulties of navigating the fierce outer atmosphere of the Earth.
[00:01:12.000 --> 00:01:18.000] These astronauts knew the dangers, and they faced them willingly,
[00:01:18.000 --> 00:01:23.000] knowing they had a high and noble purpose in life.
[00:01:23.000 --> 00:01:31.000] Because of their courage and daring and idealism, we will miss them all the more.
[00:01:31.000 --> 00:01:36.000] All Americans today are thinking as well of the families of these men and women
[00:01:36.000 --> 00:01:40.000] who have been given this sudden shock and grief.
[00:01:40.000 --> 00:01:45.000] You're not alone. Our entire nation grieves with you,
[00:01:45.000 --> 00:01:52.000] and those you love will always have the respect and gratitude of this country.
[00:01:52.000 --> 00:01:56.000] The cause in which they died will continue.
[00:01:56.000 --> 00:02:04.000] Mankind is led into the darkness beyond our world by the inspiration of discovery
[00:02:04.000 --> 00:02:11.000] and the longing to understand. Our journey into space will go on.
[00:02:11.000 --> 00:02:16.000] In the skies today, we saw destruction and tragedy.
[00:02:16.000 --> 00:02:22.000] Yet farther than we can see, there is comfort and hope.
[00:02:22.000 --> 00:02:29.000] In the words of the prophet Isaiah, "Lift your eyes and look to the heavens
[00:02:29.000 --> 00:02:35.000] who created all these. He who brings out the starry hosts one by one
[00:02:35.000 --> 00:02:39.000] and calls them each by name."
[00:02:39.000 --> 00:02:46.000] Because of His great power and mighty strength, not one of them is missing.
[00:02:46.000 --> 00:02:55.000] The same Creator who names the stars also knows the names of the seven souls we mourn today.
[00:02:55.000 --> 00:03:01.000] The crew of the shuttle Columbia did not return safely to earth,
[00:03:01.000 --> 00:03:05.000] yet we can pray that all are safely home.
[00:03:05.000 --> 00:03:13.000] May God bless the grieving families, and may God continue to bless America.
[00:03:13.000 --> 00:03:19.000] [Silence]
whisper_print_timings: fallbacks = 1 p / 0 h
whisper_print_timings: load time = 569.03 ms
whisper_print_timings: mel time = 146.85 ms
whisper_print_timings: sample time = 238.66 ms / 553 runs ( 0.43 ms per run)
whisper_print_timings: encode time = 18665.10 ms / 9 runs ( 2073.90 ms per run)
whisper_print_timings: decode time = 13090.93 ms / 549 runs ( 23.85 ms per run)
whisper_print_timings: total time = 32733.52 ms
whisper_model_load: ggml ctx size = 1644.97 MB
whisper_model_load: memory size = 182.62 MB
whisper_model_load: model size = 1462.12 MB
main: processing 'samples/gb1.wav' (3179750 samples, 198.7 sec), 8 threads, lang = en, task = transcribe, timestamps = 1 ...
[00:00.000 --> 00:08.000] My fellow Americans, this day has brought terrible news and great sadness to our country.
[00:08.000 --> 00:17.000] At nine o'clock this morning, Mission Control in Houston lost contact with our Space Shuttle Columbia.
[00:17.000 --> 00:23.000] A short time later, debris was seen falling from the skies above Texas.
[00:23.000 --> 00:29.000] The Columbia's lost. There are no survivors.
[00:29.000 --> 00:32.000] On board was a crew of seven.
[00:32.000 --> 00:39.000] Colonel Rick Husband, Lieutenant Colonel Michael Anderson, Commander Laurel Clark,
[00:39.000 --> 00:48.000] Captain David Brown, Commander William McCool, Dr. Kultna Shavla, and Ilan Ramon,
[00:48.000 --> 00:52.000] a colonel in the Israeli Air Force.
[00:52.000 --> 00:58.000] These men and women assumed great risk in the service to all humanity.
[00:58.000 --> 01:03.000] In an age when space flight has come to seem almost routine,
[01:03.000 --> 01:07.000] it is easy to overlook the dangers of travel by rocket
[01:07.000 --> 01:12.000] and the difficulties of navigating the fierce outer atmosphere of the Earth.
[01:12.000 --> 01:18.000] These astronauts knew the dangers, and they faced them willingly,
[01:18.000 --> 01:23.000] knowing they had a high and noble purpose in life.
[01:23.000 --> 01:31.000] Because of their courage and daring and idealism, we will miss them all the more.
[01:31.000 --> 01:36.000] All Americans today are thinking as well of the families of these men and women
[01:36.000 --> 01:40.000] who have been given this sudden shock and grief.
[01:40.000 --> 01:45.000] You're not alone. Our entire nation grieves with you,
[01:45.000 --> 01:52.000] and those you love will always have the respect and gratitude of this country.
[01:52.000 --> 01:56.000] The cause in which they died will continue.
[01:56.000 --> 02:04.000] Mankind is led into the darkness beyond our world by the inspiration of discovery
[02:04.000 --> 02:11.000] and the longing to understand. Our journey into space will go on.
[02:11.000 --> 02:16.000] In the skies today, we saw destruction and tragedy.
[02:16.000 --> 02:22.000] Yet farther than we can see, there is comfort and hope.
[02:22.000 --> 02:29.000] In the words of the prophet Isaiah, "Lift your eyes and look to the heavens
[02:29.000 --> 02:35.000] who created all these. He who brings out the starry hosts one by one
[02:35.000 --> 02:39.000] and calls them each by name."
[02:39.000 --> 02:46.000] Because of His great power and mighty strength, not one of them is missing.
[02:46.000 --> 02:55.000] The same Creator who names the stars also knows the names of the seven souls we mourn today.
[02:55.000 --> 03:01.000] The crew of the shuttle Columbia did not return safely to earth,
[03:01.000 --> 03:05.000] yet we can pray that all are safely home.
[03:05.000 --> 03:13.000] May God bless the grieving families, and may God continue to bless America.
[03:13.000 --> 03:41.000] Audio
whisper_print_timings: load time = 575.92 ms
whisper_print_timings: mel time = 230.60 ms
whisper_print_timings: sample time = 73.19 ms
whisper_print_timings: encode time = 19552.61 ms / 814.69 ms per layer
whisper_print_timings: decode time = 13249.96 ms / 552.08 ms per layer
whisper_print_timings: total time = 33686.27 ms
```
</details>
@ -332,14 +321,14 @@ to highlight words with high or low confidence:
## Controlling the length of the generated text segments (experimental)
For example, to limit the line length to a maximum of 16 characters, simply add `-ml 16`:
For example, to limit the line length to a maximum of 16 characters, simply add `-ml 16`:
```java
./main -m ./models/ggml-base.en.bin -f ./samples/jfk.wav -ml 16
whisper_model_load: loading model from './models/ggml-base.en.bin'
...
system_info: n_threads = 4 / 10 | AVX2 = 0 | AVX512 = 0 | NEON = 1 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 |
system_info: n_threads = 4 / 10 | AVX2 = 0 | AVX512 = 0 | NEON = 1 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 |
main: processing './samples/jfk.wav' (176000 samples, 11.0 sec), 4 threads, 1 processors, lang = en, task = transcribe, timestamps = 1 ...
@ -363,7 +352,7 @@ The `--max-len` argument can be used to obtain word-level timestamps. Simply use
whisper_model_load: loading model from './models/ggml-base.en.bin'
...
system_info: n_threads = 4 / 10 | AVX2 = 0 | AVX512 = 0 | NEON = 1 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 |
system_info: n_threads = 4 / 10 | AVX2 = 0 | AVX512 = 0 | NEON = 1 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 |
main: processing './samples/jfk.wav' (176000 samples, 11.0 sec), 4 threads, 1 processors, lang = en, task = transcribe, timestamps = 1 ...
@ -433,19 +422,6 @@ https://user-images.githubusercontent.com/1991296/199337538-b7b0c7a3-2753-4a88-a
---
## Video comparison of different models
Use the [extra/bench-wts.sh](https://github.com/ggerganov/whisper.cpp/blob/master/extra/bench-wts.sh) script to generate a video in the following format:
```java
./extra/bench-wts.sh samples/jfk.wav
ffplay ./samples/jfk.wav.all.mp4
```
https://user-images.githubusercontent.com/1991296/223206245-2d36d903-cf8e-4f09-8c3b-eb9f9c39d6fc.mp4
---
## Benchmarks
In order to have an objective comparison of the performance of the inference across different system configurations,
@ -477,14 +453,8 @@ in [models](models).
- [X] Rust: [tazz4843/whisper-rs](https://github.com/tazz4843/whisper-rs) | [#310](https://github.com/ggerganov/whisper.cpp/discussions/310)
- [X] Javascript: [bindings/javascript](bindings/javascript) | [#309](https://github.com/ggerganov/whisper.cpp/discussions/309)
- [X] Go: [bindings/go](bindings/go) | [#312](https://github.com/ggerganov/whisper.cpp/discussions/312)
- [X] Ruby: [bindings/ruby](bindings/ruby) | [#507](https://github.com/ggerganov/whisper.cpp/discussions/507)
- [X] Objective-C / Swift: [ggerganov/whisper.spm](https://github.com/ggerganov/whisper.spm) | [#313](https://github.com/ggerganov/whisper.cpp/discussions/313)
- [X] .NET: | [#422](https://github.com/ggerganov/whisper.cpp/discussions/422)
- [sandrohanea/whisper.net](https://github.com/sandrohanea/whisper.net)
- [NickDarvey/whisper](https://github.com/NickDarvey/whisper)
- [X] Python: | [#9](https://github.com/ggerganov/whisper.cpp/issues/9)
- [stlukey/whispercpp.py](https://github.com/stlukey/whispercpp.py) (Cython)
- [aarnphm/whispercpp](https://github.com/aarnphm/whispercpp) (Pybind11)
- [ ] Python: soon | [WIP](https://github.com/ggerganov/whisper.cpp/issues/9)
## Examples

@ -25,8 +25,6 @@ func Process(model whisper.Model, path string, flags *Flags) error {
return err
}
fmt.Printf("\n%s\n", context.SystemInfo())
// Open the file
fmt.Fprintf(flags.Output(), "Loading %q\n", path)
fh, err := os.Open(path)
@ -66,13 +64,10 @@ func Process(model whisper.Model, path string, flags *Flags) error {
// Process the data
fmt.Fprintf(flags.Output(), " ...processing %q\n", path)
context.ResetTimings()
if err := context.Process(data, cb); err != nil {
return err
}
context.PrintTimings()
// Print out the results
switch {
case flags.GetOut() == "srt":

@ -49,10 +49,6 @@ func (p *Params) SetSpeedup(v bool) {
// Set language id
func (p *Params) SetLanguage(lang int) error {
if lang == -1 {
p.language = nil
return nil
}
str := C.whisper_lang_str(C.int(lang))
if str == nil {
return ErrInvalidLanguage
@ -70,11 +66,6 @@ func (p *Params) Language() int {
return int(C.whisper_lang_id(p.language))
}
// Threads available
func (p *Params) Threads() int {
return int(p.n_threads)
}
// Set number of threads to use
func (p *Params) SetThreads(threads int) {
p.n_threads = C.int(threads)

@ -1,9 +1,7 @@
package whisper
import (
"fmt"
"io"
"runtime"
"strings"
"time"
@ -46,10 +44,7 @@ func (context *context) SetLanguage(lang string) error {
if !context.model.IsMultilingual() {
return ErrModelNotMultilingual
}
if lang == "auto" {
context.params.SetLanguage(-1)
} else if id := context.model.ctx.Whisper_lang_id(lang); id < 0 {
if id := context.model.ctx.Whisper_lang_id(lang); id < 0 {
return ErrUnsupportedLanguage
} else if err := context.params.SetLanguage(id); err != nil {
return err
@ -64,10 +59,6 @@ func (context *context) IsMultilingual() bool {
// Get language
func (context *context) Language() string {
id := context.params.Language()
if id == -1 {
return "auto"
}
return whisper.Whisper_lang_str(context.params.Language())
}
@ -116,36 +107,6 @@ func (context *context) SetMaxTokensPerSegment(n uint) {
context.params.SetMaxTokensPerSegment(int(n))
}
// ResetTimings resets the mode timings. Should be called before processing
func (context *context) ResetTimings() {
context.model.ctx.Whisper_reset_timings()
}
// PrintTimings prints the model timings to stdout.
func (context *context) PrintTimings() {
context.model.ctx.Whisper_print_timings()
}
// SystemInfo returns the system information
func (context *context) SystemInfo() string {
return fmt.Sprintf("system_info: n_threads = %d / %d | %s\n",
context.params.Threads(),
runtime.NumCPU(),
whisper.Whisper_print_system_info(),
)
}
// Use mel data at offset_ms to try and auto-detect the spoken language
// Make sure to call whisper_pcm_to_mel() or whisper_set_mel() first.
// Returns the probabilities of all languages.
func (context *context) WhisperLangAutoDetect(offset_ms int, n_threads int) ([]float32, error) {
langProbs, err := context.model.ctx.Whisper_lang_auto_detect(offset_ms, n_threads)
if err != nil {
return nil, err
}
return langProbs, nil
}
// Process new sample data and return any errors
func (context *context) Process(data []float32, cb SegmentCallback) error {
if context.model.ctx == nil {

@ -29,7 +29,7 @@ type Model interface {
// Context is the speach recognition context.
type Context interface {
SetLanguage(string) error // Set the language to use for speech recognition, use "auto" for auto detect language.
SetLanguage(string) error // Set the language to use for speech recognition.
SetTranslate(bool) // Set translate flag
IsMultilingual() bool // Return true if the model is multilingual.
Language() string // Get language
@ -60,12 +60,6 @@ type Context interface {
IsNOT(Token) bool // Test for "No timestamps" token
IsLANG(Token, string) bool // Test for token associated with a specific language
IsText(Token) bool // Test for text token
// Timings
PrintTimings()
ResetTimings()
SystemInfo() string
}
// Segment is the text result of a speech recognition.

@ -94,7 +94,6 @@ func (model *model) NewContext() (Context, error) {
params.SetPrintRealtime(false)
params.SetPrintTimestamps(false)
params.SetThreads(runtime.NumCPU())
params.SetNoContext(true)
// Return new context
return newContext(model, params)

@ -20,7 +20,7 @@ extern bool callEncoderBegin(void* user_data);
// Text segment callback
// Called on every newly generated text segment
// Use the whisper_full_...() functions to obtain the text segments
static void whisper_new_segment_cb(struct whisper_context* ctx, struct whisper_state* state, int n_new, void* user_data) {
static void whisper_new_segment_cb(struct whisper_context* ctx, int n_new, void* user_data) {
if(user_data != NULL && ctx != NULL) {
callNewSegment(user_data, n_new);
}
@ -29,7 +29,7 @@ static void whisper_new_segment_cb(struct whisper_context* ctx, struct whisper_s
// Encoder begin callback
// If not NULL, called before the encoder starts
// If it returns false, the computation is aborted
static bool whisper_encoder_begin_cb(struct whisper_context* ctx, struct whisper_state* state, void* user_data) {
static bool whisper_encoder_begin_cb(struct whisper_context* ctx, void* user_data) {
if(user_data != NULL && ctx != NULL) {
return callEncoderBegin(user_data);
}

@ -1 +1 @@
Subproject commit 92d4c5c9a07b726e35c20dc513532789919e00c4
Subproject commit f6334b026f63700ff86f2a942da7ef69690d10a2

@ -1,6 +1,6 @@
{
"name": "whisper.cpp",
"version": "1.2.1",
"version": "1.1.0",
"description": "Whisper speech recognition",
"main": "whisper.js",
"scripts": {

File diff suppressed because one or more lines are too long

@ -1,7 +0,0 @@
Makefile
ggml.c
ggml.h
whisper.bundle
whisper.cpp
whisper.h
dr_wav.h

@ -1,21 +0,0 @@
require 'mkmf'
system("cp #{File.join(File.dirname(__FILE__),'..','..','..','whisper.cpp')} .")
system("cp #{File.join(File.dirname(__FILE__),'..','..','..','whisper.h')} .")
system("cp #{File.join(File.dirname(__FILE__),'..','..','..','ggml.h')} .")
system("cp #{File.join(File.dirname(__FILE__),'..','..','..','ggml.c')} .")
system("cp #{File.join(File.dirname(__FILE__),'..','..','..','examples','dr_wav.h')} .")
# need to use c++ compiler flags
$CXXFLAGS << ' -std=c++11'
# Set to true when building binary gems
if enable_config('static-stdlib', false)
$LDFLAGS << ' -static-libgcc -static-libstdc++'
end
if enable_config('march-tune-native', false)
$CFLAGS << ' -march=native -mtune=native'
$CXXFLAGS << ' -march=native -mtune=native'
end
create_makefile('whisper')

@ -1,426 +0,0 @@
#include <ruby.h>
#include "ruby_whisper.h"
#define DR_WAV_IMPLEMENTATION
#include "dr_wav.h"
#include <cmath>
#include <fstream>
#include <cstdio>
#include <string>
#include <thread>
#include <vector>
#ifdef __cplusplus
extern "C" {
#endif
#define BOOL_PARAMS_SETTER(self, prop, value) \
ruby_whisper_params *rwp; \
Data_Get_Struct(self, ruby_whisper_params, rwp); \
if (value == Qfalse || value == Qnil) { \
rwp->params.prop = false; \
} else { \
rwp->params.prop = true; \
} \
return value; \
#define BOOL_PARAMS_GETTER(self, prop) \
ruby_whisper_params *rwp; \
Data_Get_Struct(self, ruby_whisper_params, rwp); \
if (rwp->params.prop) { \
return Qtrue; \
} else { \
return Qfalse; \
}
VALUE mWhisper;
VALUE cContext;
VALUE cParams;
static void ruby_whisper_free(ruby_whisper *rw) {
if (rw->context) {
whisper_free(rw->context);
rw->context = NULL;
}
}
static void ruby_whisper_params_free(ruby_whisper_params *rwp) {
}
void rb_whisper_mark(ruby_whisper *rw) {
// call rb_gc_mark on any ruby references in rw
}
void rb_whisper_free(ruby_whisper *rw) {
ruby_whisper_free(rw);
free(rw);
}
void rb_whisper_params_mark(ruby_whisper_params *rwp) {
}
void rb_whisper_params_free(ruby_whisper_params *rwp) {
ruby_whisper_params_free(rwp);
free(rwp);
}
static VALUE ruby_whisper_allocate(VALUE klass) {
ruby_whisper *rw;
rw = ALLOC(ruby_whisper);
rw->context = NULL;
return Data_Wrap_Struct(klass, rb_whisper_mark, rb_whisper_free, rw);
}
static VALUE ruby_whisper_params_allocate(VALUE klass) {
ruby_whisper_params *rwp;
rwp = ALLOC(ruby_whisper_params);
rwp->params = whisper_full_default_params(WHISPER_SAMPLING_GREEDY);
return Data_Wrap_Struct(klass, rb_whisper_params_mark, rb_whisper_params_free, rwp);
}
static VALUE ruby_whisper_initialize(int argc, VALUE *argv, VALUE self) {
ruby_whisper *rw;
VALUE whisper_model_file_path;
// TODO: we can support init from buffer here too maybe another ruby object to expose
rb_scan_args(argc, argv, "01", &whisper_model_file_path);
Data_Get_Struct(self, ruby_whisper, rw);
if (!rb_respond_to(whisper_model_file_path, rb_intern("to_s"))) {
rb_raise(rb_eRuntimeError, "Expected file path to model to initialize Whisper::Context");
}
rw->context = whisper_init_from_file(StringValueCStr(whisper_model_file_path));
if (rw->context == nullptr) {
rb_raise(rb_eRuntimeError, "error: failed to initialize whisper context");
}
return self;
}
/*
* transcribe a single file
* can emit to a block results
*
**/
static VALUE ruby_whisper_transcribe(int argc, VALUE *argv, VALUE self) {
ruby_whisper *rw;
ruby_whisper_params *rwp;
VALUE wave_file_path, blk, params;
rb_scan_args(argc, argv, "02&", &wave_file_path, &params, &blk);
Data_Get_Struct(self, ruby_whisper, rw);
Data_Get_Struct(params, ruby_whisper_params, rwp);
if (!rb_respond_to(wave_file_path, rb_intern("to_s"))) {
rb_raise(rb_eRuntimeError, "Expected file path to wave file");
}
std::string fname_inp = StringValueCStr(wave_file_path);
std::vector<float> pcmf32; // mono-channel F32 PCM
std::vector<std::vector<float>> pcmf32s; // stereo-channel F32 PCM
// WAV input - this is directly from main.cpp example
{
drwav wav;
std::vector<uint8_t> wav_data; // used for pipe input from stdin
if (fname_inp == "-") {
{
uint8_t buf[1024];
while (true) {
const size_t n = fread(buf, 1, sizeof(buf), stdin);
if (n == 0) {
break;
}
wav_data.insert(wav_data.end(), buf, buf + n);
}
}
if (drwav_init_memory(&wav, wav_data.data(), wav_data.size(), nullptr) == false) {
fprintf(stderr, "error: failed to open WAV file from stdin\n");
return self;
}
fprintf(stderr, "%s: read %zu bytes from stdin\n", __func__, wav_data.size());
} else if (drwav_init_file(&wav, fname_inp.c_str(), nullptr) == false) {
fprintf(stderr, "error: failed to open '%s' as WAV file\n", fname_inp.c_str());
return self;
}
if (wav.channels != 1 && wav.channels != 2) {
fprintf(stderr, "WAV file '%s' must be mono or stereo\n", fname_inp.c_str());
return self;
}
if (rwp->diarize && wav.channels != 2 && rwp->params.print_timestamps == false) {
fprintf(stderr, "WAV file '%s' must be stereo for diarization and timestamps have to be enabled\n", fname_inp.c_str());
return self;
}
if (wav.sampleRate != WHISPER_SAMPLE_RATE) {
fprintf(stderr, "WAV file '%s' must be %i kHz\n", fname_inp.c_str(), WHISPER_SAMPLE_RATE/1000);
return self;
}
if (wav.bitsPerSample != 16) {
fprintf(stderr, "WAV file '%s' must be 16-bit\n", fname_inp.c_str());
return self;
}
const uint64_t n = wav_data.empty() ? wav.totalPCMFrameCount : wav_data.size()/(wav.channels*wav.bitsPerSample/8);
std::vector<int16_t> pcm16;
pcm16.resize(n*wav.channels);
drwav_read_pcm_frames_s16(&wav, n, pcm16.data());
drwav_uninit(&wav);
// convert to mono, float
pcmf32.resize(n);
if (wav.channels == 1) {
for (uint64_t i = 0; i < n; i++) {
pcmf32[i] = float(pcm16[i])/32768.0f;
}
} else {
for (uint64_t i = 0; i < n; i++) {
pcmf32[i] = float(pcm16[2*i] + pcm16[2*i + 1])/65536.0f;
}
}
if (rwp->diarize) {
// convert to stereo, float
pcmf32s.resize(2);
pcmf32s[0].resize(n);
pcmf32s[1].resize(n);
for (uint64_t i = 0; i < n; i++) {
pcmf32s[0][i] = float(pcm16[2*i])/32768.0f;
pcmf32s[1][i] = float(pcm16[2*i + 1])/32768.0f;
}
}
}
{
static bool is_aborted = false; // NOTE: this should be atomic to avoid data race
rwp->params.encoder_begin_callback = [](struct whisper_context * /*ctx*/, struct whisper_state * /*state*/, void * user_data) {
bool is_aborted = *(bool*)user_data;
return !is_aborted;
};
rwp->params.encoder_begin_callback_user_data = &is_aborted;
}
if (whisper_full_parallel(rw->context, rwp->params, pcmf32.data(), pcmf32.size(), 1) != 0) {
fprintf(stderr, "failed to process audio\n");
return self;
}
const int n_segments = whisper_full_n_segments(rw->context);
VALUE output = rb_str_new2("");
for (int i = 0; i < n_segments; ++i) {
const char * text = whisper_full_get_segment_text(rw->context, i);
output = rb_str_concat(output, rb_str_new2(text));
}
VALUE idCall = rb_intern("call");
if (blk != Qnil) {
rb_funcall(blk, idCall, 1, output);
}
return self;
}
/*
* params.language = "auto" | "en", etc...
*/
static VALUE ruby_whisper_params_set_language(VALUE self, VALUE value) {
ruby_whisper_params *rwp;
Data_Get_Struct(self, ruby_whisper_params, rwp);
if (value == Qfalse || value == Qnil) {
rwp->params.language = "auto";
} else {
rwp->params.language = StringValueCStr(value);
}
return value;
}
static VALUE ruby_whisper_params_get_language(VALUE self) {
ruby_whisper_params *rwp;
Data_Get_Struct(self, ruby_whisper_params, rwp);
if (rwp->params.language) {
return rb_str_new2(rwp->params.language);
} else {
return rb_str_new2("auto");
}
}
static VALUE ruby_whisper_params_set_translate(VALUE self, VALUE value) {
BOOL_PARAMS_SETTER(self, translate, value)
}
static VALUE ruby_whisper_params_get_translate(VALUE self) {
BOOL_PARAMS_GETTER(self, translate)
}
static VALUE ruby_whisper_params_set_no_context(VALUE self, VALUE value) {
BOOL_PARAMS_SETTER(self, no_context, value)
}
static VALUE ruby_whisper_params_get_no_context(VALUE self) {
BOOL_PARAMS_GETTER(self, no_context)
}
static VALUE ruby_whisper_params_set_single_segment(VALUE self, VALUE value) {
BOOL_PARAMS_SETTER(self, single_segment, value)
}
static VALUE ruby_whisper_params_get_single_segment(VALUE self) {
BOOL_PARAMS_GETTER(self, single_segment)
}
static VALUE ruby_whisper_params_set_print_special(VALUE self, VALUE value) {
BOOL_PARAMS_SETTER(self, print_special, value)
}
static VALUE ruby_whisper_params_get_print_special(VALUE self) {
BOOL_PARAMS_GETTER(self, print_special)
}
static VALUE ruby_whisper_params_set_print_progress(VALUE self, VALUE value) {
BOOL_PARAMS_SETTER(self, print_progress, value)
}
static VALUE ruby_whisper_params_get_print_progress(VALUE self) {
BOOL_PARAMS_GETTER(self, print_progress)
}
static VALUE ruby_whisper_params_set_print_realtime(VALUE self, VALUE value) {
BOOL_PARAMS_SETTER(self, print_realtime, value)
}
static VALUE ruby_whisper_params_get_print_realtime(VALUE self) {
BOOL_PARAMS_GETTER(self, print_realtime)
}
static VALUE ruby_whisper_params_set_print_timestamps(VALUE self, VALUE value) {
BOOL_PARAMS_SETTER(self, print_timestamps, value)
}
static VALUE ruby_whisper_params_get_print_timestamps(VALUE self) {
BOOL_PARAMS_GETTER(self, print_timestamps)
}
static VALUE ruby_whisper_params_set_suppress_blank(VALUE self, VALUE value) {
BOOL_PARAMS_SETTER(self, suppress_blank, value)
}
static VALUE ruby_whisper_params_get_suppress_blank(VALUE self) {
BOOL_PARAMS_GETTER(self, suppress_blank)
}
static VALUE ruby_whisper_params_set_suppress_non_speech_tokens(VALUE self, VALUE value) {
BOOL_PARAMS_SETTER(self, suppress_non_speech_tokens, value)
}
static VALUE ruby_whisper_params_get_suppress_non_speech_tokens(VALUE self) {
BOOL_PARAMS_GETTER(self, suppress_non_speech_tokens)
}
static VALUE ruby_whisper_params_get_token_timestamps(VALUE self) {
BOOL_PARAMS_GETTER(self, token_timestamps)
}
static VALUE ruby_whisper_params_set_token_timestamps(VALUE self, VALUE value) {
BOOL_PARAMS_SETTER(self, token_timestamps, value)
}
static VALUE ruby_whisper_params_get_split_on_word(VALUE self) {
BOOL_PARAMS_GETTER(self, split_on_word)
}
static VALUE ruby_whisper_params_set_split_on_word(VALUE self, VALUE value) {
BOOL_PARAMS_SETTER(self, split_on_word, value)
}
static VALUE ruby_whisper_params_get_speed_up(VALUE self) {
BOOL_PARAMS_GETTER(self, speed_up)
}
static VALUE ruby_whisper_params_set_speed_up(VALUE self, VALUE value) {
BOOL_PARAMS_SETTER(self, speed_up, value)
}
static VALUE ruby_whisper_params_get_diarize(VALUE self) {
ruby_whisper_params *rwp;
Data_Get_Struct(self, ruby_whisper_params, rwp);
if (rwp->diarize) {
return Qtrue;
} else {
return Qfalse;
}
}
static VALUE ruby_whisper_params_set_diarize(VALUE self, VALUE value) {
ruby_whisper_params *rwp;
Data_Get_Struct(self, ruby_whisper_params, rwp);
if (value == Qfalse || value == Qnil) {
rwp->diarize = false;
} else {
rwp->diarize = true;
} \
return value;
}
static VALUE ruby_whisper_params_get_offset(VALUE self) {
ruby_whisper_params *rwp;
Data_Get_Struct(self, ruby_whisper_params, rwp);
return INT2NUM(rwp->params.offset_ms);
}
static VALUE ruby_whisper_params_set_offset(VALUE self, VALUE value) {
ruby_whisper_params *rwp;
Data_Get_Struct(self, ruby_whisper_params, rwp);
rwp->params.offset_ms = NUM2INT(value);
return value;
}
static VALUE ruby_whisper_params_get_duration(VALUE self) {
ruby_whisper_params *rwp;
Data_Get_Struct(self, ruby_whisper_params, rwp);
return INT2NUM(rwp->params.duration_ms);
}
static VALUE ruby_whisper_params_set_duration(VALUE self, VALUE value) {
ruby_whisper_params *rwp;
Data_Get_Struct(self, ruby_whisper_params, rwp);
rwp->params.duration_ms = NUM2INT(value);
return value;
}
static VALUE ruby_whisper_params_get_max_text_tokens(VALUE self) {
ruby_whisper_params *rwp;
Data_Get_Struct(self, ruby_whisper_params, rwp);
return INT2NUM(rwp->params.n_max_text_ctx);
}
static VALUE ruby_whisper_params_set_max_text_tokens(VALUE self, VALUE value) {
ruby_whisper_params *rwp;
Data_Get_Struct(self, ruby_whisper_params, rwp);
rwp->params.n_max_text_ctx = NUM2INT(value);
return value;
}
void Init_whisper() {
mWhisper = rb_define_module("Whisper");
cContext = rb_define_class_under(mWhisper, "Context", rb_cObject);
cParams = rb_define_class_under(mWhisper, "Params", rb_cObject);
rb_define_alloc_func(cContext, ruby_whisper_allocate);
rb_define_method(cContext, "initialize", ruby_whisper_initialize, -1);
rb_define_method(cContext, "transcribe", ruby_whisper_transcribe, -1);
rb_define_alloc_func(cParams, ruby_whisper_params_allocate);
rb_define_method(cParams, "language=", ruby_whisper_params_set_language, 1);
rb_define_method(cParams, "language", ruby_whisper_params_get_language, 0);
rb_define_method(cParams, "translate=", ruby_whisper_params_set_translate, 1);
rb_define_method(cParams, "translate", ruby_whisper_params_get_translate, 0);
rb_define_method(cParams, "no_context=", ruby_whisper_params_set_no_context, 1);
rb_define_method(cParams, "no_context", ruby_whisper_params_get_no_context, 0);
rb_define_method(cParams, "single_segment=", ruby_whisper_params_set_single_segment, 1);
rb_define_method(cParams, "single_segment", ruby_whisper_params_get_single_segment, 0);
rb_define_method(cParams, "print_special", ruby_whisper_params_get_print_special, 0);
rb_define_method(cParams, "print_special=", ruby_whisper_params_set_print_special, 1);
rb_define_method(cParams, "print_progress", ruby_whisper_params_get_print_progress, 0);
rb_define_method(cParams, "print_progress=", ruby_whisper_params_set_print_progress, 1);
rb_define_method(cParams, "print_realtime", ruby_whisper_params_get_print_realtime, 0);
rb_define_method(cParams, "print_realtime=", ruby_whisper_params_set_print_realtime, 1);
rb_define_method(cParams, "print_timestamps", ruby_whisper_params_get_print_timestamps, 0);
rb_define_method(cParams, "print_timestamps=", ruby_whisper_params_set_print_timestamps, 1);
rb_define_method(cParams, "suppress_blank", ruby_whisper_params_get_suppress_blank, 0);
rb_define_method(cParams, "suppress_blank=", ruby_whisper_params_set_suppress_blank, 1);
rb_define_method(cParams, "suppress_non_speech_tokens", ruby_whisper_params_get_suppress_non_speech_tokens, 0);
rb_define_method(cParams, "suppress_non_speech_tokens=", ruby_whisper_params_set_suppress_non_speech_tokens, 1);
rb_define_method(cParams, "token_timestamps", ruby_whisper_params_get_token_timestamps, 0);
rb_define_method(cParams, "token_timestamps=", ruby_whisper_params_set_token_timestamps, 1);
rb_define_method(cParams, "split_on_word", ruby_whisper_params_get_split_on_word, 0);
rb_define_method(cParams, "split_on_word=", ruby_whisper_params_set_split_on_word, 1);
rb_define_method(cParams, "speed_up", ruby_whisper_params_get_speed_up, 0);
rb_define_method(cParams, "speed_up=", ruby_whisper_params_set_speed_up, 1);
rb_define_method(cParams, "diarize", ruby_whisper_params_get_diarize, 0);
rb_define_method(cParams, "diarize=", ruby_whisper_params_set_diarize, 1);
rb_define_method(cParams, "offset", ruby_whisper_params_get_offset, 0);
rb_define_method(cParams, "offset=", ruby_whisper_params_set_offset, 1);
rb_define_method(cParams, "duration", ruby_whisper_params_get_duration, 0);
rb_define_method(cParams, "duration=", ruby_whisper_params_set_duration, 1);
rb_define_method(cParams, "max_text_tokens", ruby_whisper_params_get_max_text_tokens, 0);
rb_define_method(cParams, "max_text_tokens=", ruby_whisper_params_set_max_text_tokens, 1);
}
#ifdef __cplusplus
}
#endif

@ -1,15 +0,0 @@
#ifndef __RUBY_WHISPER_H
#define __RUBY_WHISPER_H
#include "whisper.h"
typedef struct {
struct whisper_context *context;
} ruby_whisper;
typedef struct {
struct whisper_full_params params;
bool diarize;
} ruby_whisper_params;
#endif

@ -1,138 +0,0 @@
TOPDIR = File.expand_path(File.join(File.dirname(__FILE__), '..'))
EXTDIR = File.join(TOPDIR, 'ext')
#$LIBDIR = File.join(TOPDIR, 'lib')
#$:.unshift(LIBDIR)
$:.unshift(EXTDIR)
require 'whisper'
require 'test/unit'
class TestWhisper < Test::Unit::TestCase
def setup
@params = Whisper::Params.new
end
def test_language
@params.language = "en"
assert_equal @params.language, "en"
@params.language = "auto"
assert_equal @params.language, "auto"
end
def test_offset
@params.offset = 10_000
assert_equal @params.offset, 10_000
@params.offset = 0
assert_equal @params.offset, 0
end
def test_duration
@params.duration = 60_000
assert_equal @params.duration, 60_000
@params.duration = 0
assert_equal @params.duration, 0
end
def test_max_text_tokens
@params.max_text_tokens = 300
assert_equal @params.max_text_tokens, 300
@params.max_text_tokens = 0
assert_equal @params.max_text_tokens, 0
end
def test_translate
@params.translate = true
assert @params.translate
@params.translate = false
assert !@params.translate
end
def test_no_context
@params.no_context = true
assert @params.no_context
@params.no_context = false
assert !@params.no_context
end
def test_single_segment
@params.single_segment = true
assert @params.single_segment
@params.single_segment = false
assert !@params.single_segment
end
def test_print_special
@params.print_special = true
assert @params.print_special
@params.print_special = false
assert !@params.print_special
end
def test_print_progress
@params.print_progress = true
assert @params.print_progress
@params.print_progress = false
assert !@params.print_progress
end
def test_print_realtime
@params.print_realtime = true
assert @params.print_realtime
@params.print_realtime = false
assert !@params.print_realtime
end
def test_print_timestamps
@params.print_timestamps = true
assert @params.print_timestamps
@params.print_timestamps = false
assert !@params.print_timestamps
end
def test_suppress_blank
@params.suppress_blank = true
assert @params.suppress_blank
@params.suppress_blank = false
assert !@params.suppress_blank
end
def test_suppress_non_speech_tokens
@params.suppress_non_speech_tokens = true
assert @params.suppress_non_speech_tokens
@params.suppress_non_speech_tokens = false
assert !@params.suppress_non_speech_tokens
end
def test_token_timestamps
@params.token_timestamps = true
assert @params.token_timestamps
@params.token_timestamps = false
assert !@params.token_timestamps
end
def test_split_on_word
@params.split_on_word = true
assert @params.split_on_word
@params.split_on_word = false
assert !@params.split_on_word
end
def test_speed_up
@params.speed_up = true
assert @params.speed_up
@params.speed_up = false
assert !@params.speed_up
end
def test_whisper
@whisper = Whisper::Context.new(File.join(TOPDIR, '..', '..', 'models', 'ggml-base.en.bin'))
params = Whisper::Params.new
params.print_timestamps = false
jfk = File.join(TOPDIR, '..', '..', 'samples', 'jfk.wav')
@whisper.transcribe(jfk, params) {|text|
assert_match /ask not what your country can do for you, ask what you can do for your country/, text
}
end
end

@ -14,37 +14,6 @@ if (WHISPER_SUPPORT_SDL2)
message(STATUS "SDL2_LIBRARIES = ${SDL2_LIBRARIES}")
endif()
# common
set(TARGET common)
add_library(${TARGET} STATIC
common.h
common.cpp
)
include(DefaultTargetOptions)
set_target_properties(${TARGET} PROPERTIES POSITION_INDEPENDENT_CODE ON)
if (WHISPER_SUPPORT_SDL2)
# common-sdl
set(TARGET common-sdl)
add_library(${TARGET} STATIC
common-sdl.h
common-sdl.cpp
)
include(DefaultTargetOptions)
target_include_directories(${TARGET} PUBLIC ${SDL2_INCLUDE_DIRS})
target_link_libraries(${TARGET} PRIVATE ${SDL2_LIBRARIES})
set_target_properties(${TARGET} PROPERTIES POSITION_INDEPENDENT_CODE ON)
endif()
# examples
include_directories(${CMAKE_CURRENT_SOURCE_DIR})
@ -55,8 +24,6 @@ if (EMSCRIPTEN)
add_subdirectory(command.wasm)
add_subdirectory(talk.wasm)
add_subdirectory(bench.wasm)
elseif(CMAKE_JS_VERSION)
add_subdirectory(addon.node)
else()
add_subdirectory(main)
add_subdirectory(stream)

@ -1,3 +0,0 @@
.idea
node_modules
build

@ -1,31 +0,0 @@
set(TARGET whisper-addon)
# Base settings
#==================================================================
# env var supported by cmake-js
add_definitions(-DNAPI_VERSION=4)
include_directories(${CMAKE_JS_INC})
#==================================================================
add_library(${TARGET} SHARED ${CMAKE_JS_SRC} addon.cpp)
set_target_properties(${TARGET} PROPERTIES PREFIX "" SUFFIX ".node")
include(DefaultTargetOptions)
# Include N-API wrappers
#==================================================================
execute_process(COMMAND node -p "require('node-addon-api').include"
WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
OUTPUT_VARIABLE NODE_ADDON_API_DIR
)
string(REPLACE "\n" "" NODE_ADDON_API_DIR ${NODE_ADDON_API_DIR})
string(REPLACE "\"" "" NODE_ADDON_API_DIR ${NODE_ADDON_API_DIR})
target_include_directories(${TARGET} PRIVATE ${NODE_ADDON_API_DIR})
#==================================================================
target_link_libraries(${TARGET} ${CMAKE_JS_LIB} common whisper ${CMAKE_THREAD_LIBS_INIT})
if(MSVC AND CMAKE_JS_NODELIB_DEF AND CMAKE_JS_NODELIB_TARGET)
# Generate node.lib
execute_process(COMMAND ${CMAKE_AR} /def:${CMAKE_JS_NODELIB_DEF} /out:${CMAKE_JS_NODELIB_TARGET} ${CMAKE_STATIC_LINKER_FLAGS})
endif()

@ -1,37 +0,0 @@
# addon
This is an addon demo that can **perform whisper model reasoning in `node` and `electron` environments**, based on [cmake-js](https://github.com/cmake-js/cmake-js).
It can be used as a reference for using the whisper.cpp project in other node projects.
## Install
```shell
npm install
```
## Compile
Make sure it is in the project root directory and compiled with make-js.
```shell
npx cmake-js compile -T whisper-addon -B Release
```
For Electron addon and cmake-js options, you can see [cmake-js](https://github.com/cmake-js/cmake-js) and make very few configuration changes.
> Such as appointing special cmake path:
> ```shell
> npx cmake-js compile -c 'xxx/cmake' -T whisper-addon -B Release
> ```
## Run
```shell
cd examples/addon.node
node index.js --language='language' --model='model-path' --fname_inp='file-path'
```
Because this is a simple Demo, only the above parameters are set in the node environment.
Other parameters can also be specified in the node environment.

@ -1,15 +0,0 @@
const path = require('path');
const { whisper } = require(path.join(__dirname, '../../../build/Release/whisper-addon'));
const whisperParamsMock = {
language: 'en',
model: path.join(__dirname, '../../../models/ggml-base.en.bin'),
fname_inp: path.join(__dirname, '../../../samples/jfk.wav'),
};
describe("Run whisper.node", () => {
test("it should receive a non-empty value", () => {
expect(whisper(whisperParamsMock).length).toBeGreaterThan(0);
});
});

@ -1,342 +0,0 @@
#include "napi.h"
#include "common.h"
#include "whisper.h"
#include <string>
#include <thread>
#include <vector>
#include <cmath>
#include <cstdint>
struct whisper_params {
int32_t n_threads = std::min(4, (int32_t) std::thread::hardware_concurrency());
int32_t n_processors = 1;
int32_t offset_t_ms = 0;
int32_t offset_n = 0;
int32_t duration_ms = 0;
int32_t max_context = -1;
int32_t max_len = 0;
int32_t best_of = 5;
int32_t beam_size = -1;
float word_thold = 0.01f;
float entropy_thold = 2.4f;
float logprob_thold = -1.0f;
bool speed_up = false;
bool translate = false;
bool diarize = false;
bool output_txt = false;
bool output_vtt = false;
bool output_srt = false;
bool output_wts = false;
bool output_csv = false;
bool print_special = false;
bool print_colors = false;
bool print_progress = false;
bool no_timestamps = false;
std::string language = "en";
std::string prompt;
std::string model = "../../ggml-large.bin";
std::vector<std::string> fname_inp = {};
std::vector<std::string> fname_out = {};
};
struct whisper_print_user_data {
const whisper_params * params;
const std::vector<std::vector<float>> * pcmf32s;
};
// 500 -> 00:05.000
// 6000 -> 01:00.000
std::string to_timestamp(int64_t t, bool comma = false) {
int64_t msec = t * 10;
int64_t hr = msec / (1000 * 60 * 60);
msec = msec - hr * (1000 * 60 * 60);
int64_t min = msec / (1000 * 60);
msec = msec - min * (1000 * 60);
int64_t sec = msec / 1000;
msec = msec - sec * 1000;
char buf[32];
snprintf(buf, sizeof(buf), "%02d:%02d:%02d%s%03d", (int) hr, (int) min, (int) sec, comma ? "," : ".", (int) msec);
return std::string(buf);
}
int timestamp_to_sample(int64_t t, int n_samples) {
return std::max(0, std::min((int) n_samples - 1, (int) ((t*WHISPER_SAMPLE_RATE)/100)));
}
void whisper_print_segment_callback(struct whisper_context * ctx, struct whisper_state * state, int n_new, void * user_data) {
const auto & params = *((whisper_print_user_data *) user_data)->params;
const auto & pcmf32s = *((whisper_print_user_data *) user_data)->pcmf32s;
const int n_segments = whisper_full_n_segments(ctx);
std::string speaker = "";
int64_t t0;
int64_t t1;
// print the last n_new segments
const int s0 = n_segments - n_new;
if (s0 == 0) {
printf("\n");
}
for (int i = s0; i < n_segments; i++) {
if (!params.no_timestamps || params.diarize) {
t0 = whisper_full_get_segment_t0(ctx, i);
t1 = whisper_full_get_segment_t1(ctx, i);
}
if (!params.no_timestamps) {
printf("[%s --> %s] ", to_timestamp(t0).c_str(), to_timestamp(t1).c_str());
}
if (params.diarize && pcmf32s.size() == 2) {
const int64_t n_samples = pcmf32s[0].size();
const int64_t is0 = timestamp_to_sample(t0, n_samples);
const int64_t is1 = timestamp_to_sample(t1, n_samples);
double energy0 = 0.0f;
double energy1 = 0.0f;
for (int64_t j = is0; j < is1; j++) {
energy0 += fabs(pcmf32s[0][j]);
energy1 += fabs(pcmf32s[1][j]);
}
if (energy0 > 1.1*energy1) {
speaker = "(speaker 0)";
} else if (energy1 > 1.1*energy0) {
speaker = "(speaker 1)";
} else {
speaker = "(speaker ?)";
}
//printf("is0 = %lld, is1 = %lld, energy0 = %f, energy1 = %f, %s\n", is0, is1, energy0, energy1, speaker.c_str());
}
// colorful print bug
//
const char * text = whisper_full_get_segment_text(ctx, i);
printf("%s%s", speaker.c_str(), text);
// with timestamps or speakers: each segment on new line
if (!params.no_timestamps || params.diarize) {
printf("\n");
}
fflush(stdout);
}
}
int run(whisper_params &params, std::vector<std::vector<std::string>> &result) {
if (params.fname_inp.empty()) {
fprintf(stderr, "error: no input files specified\n");
return 2;
}
if (params.language != "auto" && whisper_lang_id(params.language.c_str()) == -1) {
fprintf(stderr, "error: unknown language '%s'\n", params.language.c_str());
exit(0);
}
// whisper init
struct whisper_context * ctx = whisper_init_from_file(params.model.c_str());
if (ctx == nullptr) {
fprintf(stderr, "error: failed to initialize whisper context\n");
return 3;
}
// initial prompt
std::vector<whisper_token> prompt_tokens;
if (!params.prompt.empty()) {
prompt_tokens.resize(1024);
prompt_tokens.resize(whisper_tokenize(ctx, params.prompt.c_str(), prompt_tokens.data(), prompt_tokens.size()));
fprintf(stderr, "\n");
fprintf(stderr, "initial prompt: '%s'\n", params.prompt.c_str());
fprintf(stderr, "initial tokens: [ ");
for (int i = 0; i < (int) prompt_tokens.size(); ++i) {
fprintf(stderr, "%d ", prompt_tokens[i]);
}
fprintf(stderr, "]\n");
}
for (int f = 0; f < (int) params.fname_inp.size(); ++f) {
const auto fname_inp = params.fname_inp[f];
const auto fname_out = f < (int)params.fname_out.size() && !params.fname_out[f].empty() ? params.fname_out[f] : params.fname_inp[f];
std::vector<float> pcmf32; // mono-channel F32 PCM
std::vector<std::vector<float>> pcmf32s; // stereo-channel F32 PCM
if (!::read_wav(fname_inp, pcmf32, pcmf32s, params.diarize)) {
fprintf(stderr, "error: failed to read WAV file '%s'\n", fname_inp.c_str());
continue;
}
// print system information
{
fprintf(stderr, "\n");
fprintf(stderr, "system_info: n_threads = %d / %d | %s\n",
params.n_threads*params.n_processors, std::thread::hardware_concurrency(), whisper_print_system_info());
}
// print some info about the processing
{
fprintf(stderr, "\n");
if (!whisper_is_multilingual(ctx)) {
if (params.language != "en" || params.translate) {
params.language = "en";
params.translate = false;
fprintf(stderr, "%s: WARNING: model is not multilingual, ignoring language and translation options\n", __func__);
}
}
fprintf(stderr, "%s: processing '%s' (%d samples, %.1f sec), %d threads, %d processors, lang = %s, task = %s, timestamps = %d ...\n",
__func__, fname_inp.c_str(), int(pcmf32.size()), float(pcmf32.size())/WHISPER_SAMPLE_RATE,
params.n_threads, params.n_processors,
params.language.c_str(),
params.translate ? "translate" : "transcribe",
params.no_timestamps ? 0 : 1);
fprintf(stderr, "\n");
}
// run the inference
{
whisper_full_params wparams = whisper_full_default_params(WHISPER_SAMPLING_GREEDY);
wparams.strategy = params.beam_size > 1 ? WHISPER_SAMPLING_BEAM_SEARCH : WHISPER_SAMPLING_GREEDY;
wparams.print_realtime = false;
wparams.print_progress = params.print_progress;
wparams.print_timestamps = !params.no_timestamps;
wparams.print_special = params.print_special;
wparams.translate = params.translate;
wparams.language = params.language.c_str();
wparams.n_threads = params.n_threads;
wparams.n_max_text_ctx = params.max_context >= 0 ? params.max_context : wparams.n_max_text_ctx;
wparams.offset_ms = params.offset_t_ms;
wparams.duration_ms = params.duration_ms;
wparams.token_timestamps = params.output_wts || params.max_len > 0;
wparams.thold_pt = params.word_thold;
wparams.entropy_thold = params.entropy_thold;
wparams.logprob_thold = params.logprob_thold;
wparams.max_len = params.output_wts && params.max_len == 0 ? 60 : params.max_len;
wparams.speed_up = params.speed_up;
wparams.greedy.best_of = params.best_of;
wparams.beam_search.beam_size = params.beam_size;
wparams.prompt_tokens = prompt_tokens.empty() ? nullptr : prompt_tokens.data();
wparams.prompt_n_tokens = prompt_tokens.empty() ? 0 : prompt_tokens.size();
whisper_print_user_data user_data = { &params, &pcmf32s };
// this callback is called on each new segment
if (!wparams.print_realtime) {
wparams.new_segment_callback = whisper_print_segment_callback;
wparams.new_segment_callback_user_data = &user_data;
}
// example for abort mechanism
// in this example, we do not abort the processing, but we could if the flag is set to true
// the callback is called before every encoder run - if it returns false, the processing is aborted
{
static bool is_aborted = false; // NOTE: this should be atomic to avoid data race
wparams.encoder_begin_callback = [](struct whisper_context * /*ctx*/, struct whisper_state * /*state*/, void * user_data) {
bool is_aborted = *(bool*)user_data;
return !is_aborted;
};
wparams.encoder_begin_callback_user_data = &is_aborted;
}
if (whisper_full_parallel(ctx, wparams, pcmf32.data(), pcmf32.size(), params.n_processors) != 0) {
fprintf(stderr, "failed to process audio\n");
return 10;
}
}
}
const int n_segments = whisper_full_n_segments(ctx);
result.resize(n_segments);
for (int i = 0; i < n_segments; ++i) {
const char * text = whisper_full_get_segment_text(ctx, i);
const int64_t t0 = whisper_full_get_segment_t0(ctx, i);
const int64_t t1 = whisper_full_get_segment_t1(ctx, i);
result[i].emplace_back(to_timestamp(t0, true));
result[i].emplace_back(to_timestamp(t1, true));
result[i].emplace_back(text);
}
whisper_print_timings(ctx);
whisper_free(ctx);
return 0;
}
Napi::Object whisper(const Napi::CallbackInfo& info) {
Napi::Env env = info.Env();
if (info.Length() <= 0 || !info[0].IsObject()) {
Napi::TypeError::New(env, "object expected").ThrowAsJavaScriptException();
}
whisper_params params;
std::vector<std::vector<std::string>> result;
Napi::Object whisper_params = info[0].As<Napi::Object>();
std::string language = whisper_params.Get("language").As<Napi::String>();
std::string model = whisper_params.Get("model").As<Napi::String>();
std::string input = whisper_params.Get("fname_inp").As<Napi::String>();
params.language = language;
params.model = model;
params.fname_inp.emplace_back(input);
// run model
run(params, result);
fprintf(stderr, "RESULT:\n");
for (auto sentence:result) {
fprintf(stderr, "t0: %s, t1: %s, content: %s \n",
sentence[0].c_str(), sentence[1].c_str(), sentence[2].c_str());
}
Napi::Object res = Napi::Array::New(env, result.size());
for (uint64_t i = 0; i < result.size(); ++i) {
Napi::Object tmp = Napi::Array::New(env, 3);
for (uint64_t j = 0; j < 3; ++j) {
tmp[j] = Napi::String::New(env, result[i][j]);
}
res[i] = tmp;
}
return res;
}
Napi::Object Init(Napi::Env env, Napi::Object exports) {
exports.Set(
Napi::String::New(env, "whisper"),
Napi::Function::New(env, whisper)
);
return exports;
}
NODE_API_MODULE(whisper, Init);

@ -1,27 +0,0 @@
const path = require('path');
const { whisper } = require(path.join(__dirname, '../../build/Release/whisper-addon'));
const whisperParams = {
language: 'en',
model: path.join(__dirname, '../../models/ggml-base.en.bin'),
fname_inp: '',
};
const arguments = process.argv.slice(2);
const params = Object.fromEntries(
arguments.reduce((pre, item) => {
if (item.startsWith("--")) {
return [...pre, item.slice(2).split("=")];
}
return pre;
}, []),
);
for (const key in params) {
if (whisperParams.hasOwnProperty(key)) {
whisperParams[key] = params[key];
}
}
console.log('whisperParams =', whisperParams);
console.log(whisper(whisperParams));

@ -1,16 +0,0 @@
{
"name": "whisper-addon",
"version": "0.0.0",
"description": "",
"main": "index.js",
"author": "Qanhe Chen",
"license": "MIT",
"scripts": {
"test": "jest"
},
"devDependencies": {
"cmake-js": "^7.1.1",
"jest": "^29.4.0",
"node-addon-api": "^5.0.0"
}
}

@ -11,7 +11,6 @@ add_executable(${TARGET}
include(DefaultTargetOptions)
target_link_libraries(${TARGET} PRIVATE
common
whisper
)

@ -1,5 +1,4 @@
#include "ggml.h"
#include "common.h"
#include "whisper.h"
#include <emscripten.h>
@ -28,6 +27,24 @@ std::string g_transcribed = "";
std::vector<float> g_pcmf32;
static std::string trim(const std::string & s) {
std::regex e("^\\s+|\\s+$");
return std::regex_replace(s, e, "");
}
static void high_pass_filter(std::vector<float> & data, float cutoff, float sample_rate) {
const float rc = 1.0f / (2.0f * M_PI * cutoff);
const float dt = 1.0f / sample_rate;
const float alpha = dt / (rc + dt);
float y = data[0];
for (size_t i = 1; i < data.size(); i++) {
y = alpha * (y + data[i] - data[i - 1]);
data[i] = y;
}
}
// compute similarity between two strings using Levenshtein distance
static float similarity(const std::string & s0, const std::string & s1) {
const size_t len0 = s0.size() + 1;
@ -58,6 +75,44 @@ void command_set_status(const std::string & status) {
g_status = status;
}
bool command_vad_simple(std::vector<float> & pcmf32, int sample_rate, int last_ms, float vad_thold, float freq_thold, bool verbose) {
const int n_samples = pcmf32.size();
const int n_samples_last = (sample_rate * last_ms) / 1000;
if (n_samples_last >= n_samples) {
// not enough samples - assume no speech
return false;
}
if (freq_thold > 0.0f) {
high_pass_filter(pcmf32, freq_thold, sample_rate);
}
float energy_all = 0.0f;
float energy_last = 0.0f;
for (size_t i = 0; i < n_samples; i++) {
energy_all += fabsf(pcmf32[i]);
if (i >= n_samples - n_samples_last) {
energy_last += fabsf(pcmf32[i]);
}
}
energy_all /= n_samples;
energy_last /= n_samples_last;
if (verbose) {
fprintf(stderr, "%s: energy_all: %f, energy_last: %f, vad_thold: %f, freq_thold: %f\n", __func__, energy_all, energy_last, vad_thold, freq_thold);
}
if (energy_last > vad_thold*energy_all) {
return false;
}
return true;
}
std::string command_transcribe(whisper_context * ctx, const whisper_full_params & wparams, const std::vector<float> & pcmf32, float & prob, int64_t & t_ms) {
const auto t_start = std::chrono::high_resolution_clock::now();
@ -100,7 +155,7 @@ void command_get_audio(int ms, int sample_rate, std::vector<float> & audio) {
const int64_t n_samples = (ms * sample_rate) / 1000;
int64_t n_take = 0;
if (n_samples > (int) g_pcmf32.size()) {
if (g_pcmf32.size() < n_samples) {
n_take = g_pcmf32.size();
} else {
n_take = n_samples;
@ -132,6 +187,7 @@ void command_main(size_t index) {
printf("command: using %d threads\n", wparams.n_threads);
bool is_running = true;
bool have_prompt = false;
bool ask_prompt = true;
bool print_energy = false;
@ -177,7 +233,7 @@ void command_main(size_t index) {
{
command_get_audio(vad_ms, WHISPER_SAMPLE_RATE, pcmf32_cur);
if (::vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1000, vad_thold, freq_thold, print_energy)) {
if (command_vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1000, vad_thold, freq_thold, print_energy)) {
fprintf(stdout, "%s: Speech detected! Processing ...\n", __func__);
command_set_status("Speech detected! Processing ...");

@ -5,5 +5,6 @@ if (WHISPER_SUPPORT_SDL2)
include(DefaultTargetOptions)
target_link_libraries(${TARGET} PRIVATE common common-sdl whisper ${CMAKE_THREAD_LIBS_INIT})
target_include_directories(${TARGET} PRIVATE ${SDL2_INCLUDE_DIRS})
target_link_libraries(${TARGET} PRIVATE whisper ${SDL2_LIBRARIES} ${CMAKE_THREAD_LIBS_INIT})
endif ()

@ -6,10 +6,11 @@
// ref: https://github.com/ggerganov/whisper.cpp/issues/171
//
#include "common.h"
#include "common-sdl.h"
#include "whisper.h"
#include <SDL.h>
#include <SDL_audio.h>
#include <sstream>
#include <cassert>
#include <cstdio>
@ -109,6 +110,309 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
fprintf(stderr, "\n");
}
//
// SDL Audio capture
//
class audio_async {
public:
audio_async(int len_ms);
~audio_async();
bool init(int capture_id, int sample_rate);
// start capturing audio via the provided SDL callback
// keep last len_ms seconds of audio in a circular buffer
bool resume();
bool pause();
bool clear();
// callback to be called by SDL
void callback(uint8_t * stream, int len);
// get audio data from the circular buffer
void get(int ms, std::vector<float> & audio);
private:
SDL_AudioDeviceID m_dev_id_in = 0;
int m_len_ms = 0;
int m_sample_rate = 0;
bool m_running = false;
std::mutex m_mutex;
std::vector<float> m_audio;
std::vector<float> m_audio_new;
size_t m_audio_pos = 0;
size_t m_audio_len = 0;
};
audio_async::audio_async(int len_ms) {
m_len_ms = len_ms;
}
audio_async::~audio_async() {
if (m_dev_id_in) {
SDL_CloseAudioDevice(m_dev_id_in);
}
}
bool audio_async::init(int capture_id, int sample_rate) {
SDL_LogSetPriority(SDL_LOG_CATEGORY_APPLICATION, SDL_LOG_PRIORITY_INFO);
if (SDL_Init(SDL_INIT_AUDIO) < 0) {
SDL_LogError(SDL_LOG_CATEGORY_APPLICATION, "Couldn't initialize SDL: %s\n", SDL_GetError());
return false;
}
SDL_SetHintWithPriority(SDL_HINT_AUDIO_RESAMPLING_MODE, "medium", SDL_HINT_OVERRIDE);
{
int nDevices = SDL_GetNumAudioDevices(SDL_TRUE);
fprintf(stderr, "%s: found %d capture devices:\n", __func__, nDevices);
for (int i = 0; i < nDevices; i++) {
fprintf(stderr, "%s: - Capture device #%d: '%s'\n", __func__, i, SDL_GetAudioDeviceName(i, SDL_TRUE));
}
}
SDL_AudioSpec capture_spec_requested;
SDL_AudioSpec capture_spec_obtained;
SDL_zero(capture_spec_requested);
SDL_zero(capture_spec_obtained);
capture_spec_requested.freq = sample_rate;
capture_spec_requested.format = AUDIO_F32;
capture_spec_requested.channels = 1;
capture_spec_requested.samples = 1024;
capture_spec_requested.callback = [](void * userdata, uint8_t * stream, int len) {
audio_async * audio = (audio_async *) userdata;
audio->callback(stream, len);
};
capture_spec_requested.userdata = this;
if (capture_id >= 0) {
fprintf(stderr, "%s: attempt to open capture device %d : '%s' ...\n", __func__, capture_id, SDL_GetAudioDeviceName(capture_id, SDL_TRUE));
m_dev_id_in = SDL_OpenAudioDevice(SDL_GetAudioDeviceName(capture_id, SDL_TRUE), SDL_TRUE, &capture_spec_requested, &capture_spec_obtained, 0);
} else {
fprintf(stderr, "%s: attempt to open default capture device ...\n", __func__);
m_dev_id_in = SDL_OpenAudioDevice(nullptr, SDL_TRUE, &capture_spec_requested, &capture_spec_obtained, 0);
}
if (!m_dev_id_in) {
fprintf(stderr, "%s: couldn't open an audio device for capture: %s!\n", __func__, SDL_GetError());
m_dev_id_in = 0;
return false;
} else {
fprintf(stderr, "%s: obtained spec for input device (SDL Id = %d):\n", __func__, m_dev_id_in);
fprintf(stderr, "%s: - sample rate: %d\n", __func__, capture_spec_obtained.freq);
fprintf(stderr, "%s: - format: %d (required: %d)\n", __func__, capture_spec_obtained.format,
capture_spec_requested.format);
fprintf(stderr, "%s: - channels: %d (required: %d)\n", __func__, capture_spec_obtained.channels,
capture_spec_requested.channels);
fprintf(stderr, "%s: - samples per frame: %d\n", __func__, capture_spec_obtained.samples);
}
m_sample_rate = capture_spec_obtained.freq;
m_audio.resize((m_sample_rate*m_len_ms)/1000);
return true;
}
bool audio_async::resume() {
if (!m_dev_id_in) {
fprintf(stderr, "%s: no audio device to resume!\n", __func__);
return false;
}
if (m_running) {
fprintf(stderr, "%s: already running!\n", __func__);
return false;
}
SDL_PauseAudioDevice(m_dev_id_in, 0);
m_running = true;
return true;
}
bool audio_async::pause() {
if (!m_dev_id_in) {
fprintf(stderr, "%s: no audio device to pause!\n", __func__);
return false;
}
if (!m_running) {
fprintf(stderr, "%s: already paused!\n", __func__);
return false;
}
SDL_PauseAudioDevice(m_dev_id_in, 1);
m_running = false;
return true;
}
bool audio_async::clear() {
if (!m_dev_id_in) {
fprintf(stderr, "%s: no audio device to clear!\n", __func__);
return false;
}
if (!m_running) {
fprintf(stderr, "%s: not running!\n", __func__);
return false;
}
{
std::lock_guard<std::mutex> lock(m_mutex);
m_audio_pos = 0;
m_audio_len = 0;
}
return true;
}
// callback to be called by SDL
void audio_async::callback(uint8_t * stream, int len) {
if (!m_running) {
return;
}
const size_t n_samples = len / sizeof(float);
m_audio_new.resize(n_samples);
memcpy(m_audio_new.data(), stream, n_samples * sizeof(float));
//fprintf(stderr, "%s: %zu samples, pos %zu, len %zu\n", __func__, n_samples, m_audio_pos, m_audio_len);
{
std::lock_guard<std::mutex> lock(m_mutex);
if (m_audio_pos + n_samples > m_audio.size()) {
const size_t n0 = m_audio.size() - m_audio_pos;
memcpy(&m_audio[m_audio_pos], stream, n0 * sizeof(float));
memcpy(&m_audio[0], &stream[n0], (n_samples - n0) * sizeof(float));
m_audio_pos = (m_audio_pos + n_samples) % m_audio.size();
m_audio_len = m_audio.size();
} else {
memcpy(&m_audio[m_audio_pos], stream, n_samples * sizeof(float));
m_audio_pos = (m_audio_pos + n_samples) % m_audio.size();
m_audio_len = std::min(m_audio_len + n_samples, m_audio.size());
}
}
}
void audio_async::get(int ms, std::vector<float> & result) {
if (!m_dev_id_in) {
fprintf(stderr, "%s: no audio device to get audio from!\n", __func__);
return;
}
if (!m_running) {
fprintf(stderr, "%s: not running!\n", __func__);
return;
}
result.clear();
{
std::lock_guard<std::mutex> lock(m_mutex);
if (ms <= 0) {
ms = m_len_ms;
}
size_t n_samples = (m_sample_rate * ms) / 1000;
if (n_samples > m_audio_len) {
n_samples = m_audio_len;
}
result.resize(n_samples);
int s0 = m_audio_pos - n_samples;
if (s0 < 0) {
s0 += m_audio.size();
}
if (s0 + n_samples > m_audio.size()) {
const size_t n0 = m_audio.size() - s0;
memcpy(result.data(), &m_audio[s0], n0 * sizeof(float));
memcpy(&result[n0], &m_audio[0], (n_samples - n0) * sizeof(float));
} else {
memcpy(result.data(), &m_audio[s0], n_samples * sizeof(float));
}
}
}
///////////////////////////
std::string trim(const std::string & s) {
std::regex e("^\\s+|\\s+$");
return std::regex_replace(s, e, "");
}
void high_pass_filter(std::vector<float> & data, float cutoff, float sample_rate) {
const float rc = 1.0f / (2.0f * M_PI * cutoff);
const float dt = 1.0f / sample_rate;
const float alpha = dt / (rc + dt);
float y = data[0];
for (size_t i = 1; i < data.size(); i++) {
y = alpha * (y + data[i] - data[i - 1]);
data[i] = y;
}
}
bool vad_simple(std::vector<float> & pcmf32, int sample_rate, int last_ms, float vad_thold, float freq_thold, bool verbose) {
const int n_samples = pcmf32.size();
const int n_samples_last = (sample_rate * last_ms) / 1000;
if (n_samples_last >= n_samples) {
// not enough samples - assume no speech
return false;
}
if (freq_thold > 0.0f) {
high_pass_filter(pcmf32, freq_thold, sample_rate);
}
float energy_all = 0.0f;
float energy_last = 0.0f;
for (int i = 0; i < n_samples; i++) {
energy_all += fabsf(pcmf32[i]);
if (i >= n_samples - n_samples_last) {
energy_last += fabsf(pcmf32[i]);
}
}
energy_all /= n_samples;
energy_last /= n_samples_last;
if (verbose) {
fprintf(stderr, "%s: energy_all: %f, energy_last: %f, vad_thold: %f, freq_thold: %f\n", __func__, energy_all, energy_last, vad_thold, freq_thold);
}
if (energy_last > vad_thold*energy_all) {
return false;
}
return true;
}
std::string transcribe(whisper_context * ctx, const whisper_params & params, const std::vector<float> & pcmf32, float & prob, int64_t & t_ms) {
const auto t_start = std::chrono::high_resolution_clock::now();
@ -198,7 +502,7 @@ std::vector<std::string> read_allowed_commands(const std::string & fname) {
std::string line;
while (std::getline(ifs, line)) {
line = ::trim(line);
line = trim(line);
if (line.empty()) {
continue;
}
@ -222,6 +526,23 @@ std::vector<std::string> get_words(const std::string &txt) {
return words;
}
// returns true if no exit event was received
bool process_sdl_events() {
SDL_Event event;
while (SDL_PollEvent(&event)) {
switch (event.type) {
case SDL_QUIT:
{
return false;
} break;
default:
break;
}
}
return true;
}
// command-list mode
// guide the transcription to match the most likely command from a provided list
int process_command_list(struct whisper_context * ctx, audio_async &audio, const whisper_params &params) {
@ -313,14 +634,14 @@ int process_command_list(struct whisper_context * ctx, audio_async &audio, const
// main loop
while (is_running) {
// handle Ctrl + C
is_running = sdl_poll_events();
is_running = process_sdl_events();
// delay
std::this_thread::sleep_for(std::chrono::milliseconds(100));
audio.get(2000, pcmf32_cur);
if (::vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1000, params.vad_thold, params.freq_thold, params.print_energy)) {
if (vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1000, params.vad_thold, params.freq_thold, params.print_energy)) {
fprintf(stdout, "%s: Speech detected! Processing ...\n", __func__);
const auto t_start = std::chrono::high_resolution_clock::now();
@ -454,7 +775,7 @@ int always_prompt_transcription(struct whisper_context * ctx, audio_async & audi
// main loop
while (is_running) {
// handle Ctrl + C
is_running = sdl_poll_events();
is_running = process_sdl_events();
// delay
std::this_thread::sleep_for(std::chrono::milliseconds(100));
@ -470,7 +791,7 @@ int always_prompt_transcription(struct whisper_context * ctx, audio_async & audi
{
audio.get(2000, pcmf32_cur);
if (::vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1000, params.vad_thold, params.freq_thold, params.print_energy)) {
if (vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1000, params.vad_thold, params.freq_thold, params.print_energy)) {
fprintf(stdout, "%s: Speech detected! Processing ...\n", __func__);
int64_t t_ms = 0;
@ -533,7 +854,7 @@ int process_general_transcription(struct whisper_context * ctx, audio_async &aud
// main loop
while (is_running) {
// handle Ctrl + C
is_running = sdl_poll_events();
is_running = process_sdl_events();
// delay
std::this_thread::sleep_for(std::chrono::milliseconds(100));
@ -549,7 +870,7 @@ int process_general_transcription(struct whisper_context * ctx, audio_async &aud
{
audio.get(2000, pcmf32_cur);
if (::vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1000, params.vad_thold, params.freq_thold, params.print_energy)) {
if (vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1000, params.vad_thold, params.freq_thold, params.print_energy)) {
fprintf(stdout, "%s: Speech detected! Processing ...\n", __func__);
int64_t t_ms = 0;

@ -1,226 +0,0 @@
#include "common-sdl.h"
audio_async::audio_async(int len_ms) {
m_len_ms = len_ms;
m_running = false;
}
audio_async::~audio_async() {
if (m_dev_id_in) {
SDL_CloseAudioDevice(m_dev_id_in);
}
}
bool audio_async::init(int capture_id, int sample_rate) {
SDL_LogSetPriority(SDL_LOG_CATEGORY_APPLICATION, SDL_LOG_PRIORITY_INFO);
if (SDL_Init(SDL_INIT_AUDIO) < 0) {
SDL_LogError(SDL_LOG_CATEGORY_APPLICATION, "Couldn't initialize SDL: %s\n", SDL_GetError());
return false;
}
SDL_SetHintWithPriority(SDL_HINT_AUDIO_RESAMPLING_MODE, "medium", SDL_HINT_OVERRIDE);
{
int nDevices = SDL_GetNumAudioDevices(SDL_TRUE);
fprintf(stderr, "%s: found %d capture devices:\n", __func__, nDevices);
for (int i = 0; i < nDevices; i++) {
fprintf(stderr, "%s: - Capture device #%d: '%s'\n", __func__, i, SDL_GetAudioDeviceName(i, SDL_TRUE));
}
}
SDL_AudioSpec capture_spec_requested;
SDL_AudioSpec capture_spec_obtained;
SDL_zero(capture_spec_requested);
SDL_zero(capture_spec_obtained);
capture_spec_requested.freq = sample_rate;
capture_spec_requested.format = AUDIO_F32;
capture_spec_requested.channels = 1;
capture_spec_requested.samples = 1024;
capture_spec_requested.callback = [](void * userdata, uint8_t * stream, int len) {
audio_async * audio = (audio_async *) userdata;
audio->callback(stream, len);
};
capture_spec_requested.userdata = this;
if (capture_id >= 0) {
fprintf(stderr, "%s: attempt to open capture device %d : '%s' ...\n", __func__, capture_id, SDL_GetAudioDeviceName(capture_id, SDL_TRUE));
m_dev_id_in = SDL_OpenAudioDevice(SDL_GetAudioDeviceName(capture_id, SDL_TRUE), SDL_TRUE, &capture_spec_requested, &capture_spec_obtained, 0);
} else {
fprintf(stderr, "%s: attempt to open default capture device ...\n", __func__);
m_dev_id_in = SDL_OpenAudioDevice(nullptr, SDL_TRUE, &capture_spec_requested, &capture_spec_obtained, 0);
}
if (!m_dev_id_in) {
fprintf(stderr, "%s: couldn't open an audio device for capture: %s!\n", __func__, SDL_GetError());
m_dev_id_in = 0;
return false;
} else {
fprintf(stderr, "%s: obtained spec for input device (SDL Id = %d):\n", __func__, m_dev_id_in);
fprintf(stderr, "%s: - sample rate: %d\n", __func__, capture_spec_obtained.freq);
fprintf(stderr, "%s: - format: %d (required: %d)\n", __func__, capture_spec_obtained.format,
capture_spec_requested.format);
fprintf(stderr, "%s: - channels: %d (required: %d)\n", __func__, capture_spec_obtained.channels,
capture_spec_requested.channels);
fprintf(stderr, "%s: - samples per frame: %d\n", __func__, capture_spec_obtained.samples);
}
m_sample_rate = capture_spec_obtained.freq;
m_audio.resize((m_sample_rate*m_len_ms)/1000);
return true;
}
bool audio_async::resume() {
if (!m_dev_id_in) {
fprintf(stderr, "%s: no audio device to resume!\n", __func__);
return false;
}
if (m_running) {
fprintf(stderr, "%s: already running!\n", __func__);
return false;
}
SDL_PauseAudioDevice(m_dev_id_in, 0);
m_running = true;
return true;
}
bool audio_async::pause() {
if (!m_dev_id_in) {
fprintf(stderr, "%s: no audio device to pause!\n", __func__);
return false;
}
if (!m_running) {
fprintf(stderr, "%s: already paused!\n", __func__);
return false;
}
SDL_PauseAudioDevice(m_dev_id_in, 1);
m_running = false;
return true;
}
bool audio_async::clear() {
if (!m_dev_id_in) {
fprintf(stderr, "%s: no audio device to clear!\n", __func__);
return false;
}
if (!m_running) {
fprintf(stderr, "%s: not running!\n", __func__);
return false;
}
{
std::lock_guard<std::mutex> lock(m_mutex);
m_audio_pos = 0;
m_audio_len = 0;
}
return true;
}
// callback to be called by SDL
void audio_async::callback(uint8_t * stream, int len) {
if (!m_running) {
return;
}
const size_t n_samples = len / sizeof(float);
m_audio_new.resize(n_samples);
memcpy(m_audio_new.data(), stream, n_samples * sizeof(float));
//fprintf(stderr, "%s: %zu samples, pos %zu, len %zu\n", __func__, n_samples, m_audio_pos, m_audio_len);
{
std::lock_guard<std::mutex> lock(m_mutex);
if (m_audio_pos + n_samples > m_audio.size()) {
const size_t n0 = m_audio.size() - m_audio_pos;
memcpy(&m_audio[m_audio_pos], stream, n0 * sizeof(float));
memcpy(&m_audio[0], &stream[n0], (n_samples - n0) * sizeof(float));
m_audio_pos = (m_audio_pos + n_samples) % m_audio.size();
m_audio_len = m_audio.size();
} else {
memcpy(&m_audio[m_audio_pos], stream, n_samples * sizeof(float));
m_audio_pos = (m_audio_pos + n_samples) % m_audio.size();
m_audio_len = std::min(m_audio_len + n_samples, m_audio.size());
}
}
}
void audio_async::get(int ms, std::vector<float> & result) {
if (!m_dev_id_in) {
fprintf(stderr, "%s: no audio device to get audio from!\n", __func__);
return;
}
if (!m_running) {
fprintf(stderr, "%s: not running!\n", __func__);
return;
}
result.clear();
{
std::lock_guard<std::mutex> lock(m_mutex);
if (ms <= 0) {
ms = m_len_ms;
}
size_t n_samples = (m_sample_rate * ms) / 1000;
if (n_samples > m_audio_len) {
n_samples = m_audio_len;
}
result.resize(n_samples);
int s0 = m_audio_pos - n_samples;
if (s0 < 0) {
s0 += m_audio.size();
}
if (s0 + n_samples > m_audio.size()) {
const size_t n0 = m_audio.size() - s0;
memcpy(result.data(), &m_audio[s0], n0 * sizeof(float));
memcpy(&result[n0], &m_audio[0], (n_samples - n0) * sizeof(float));
} else {
memcpy(result.data(), &m_audio[s0], n_samples * sizeof(float));
}
}
}
bool sdl_poll_events() {
SDL_Event event;
while (SDL_PollEvent(&event)) {
switch (event.type) {
case SDL_QUIT:
{
return false;
} break;
default:
break;
}
}
return true;
}

@ -1,50 +0,0 @@
#pragma once
#include <SDL.h>
#include <SDL_audio.h>
#include <atomic>
#include <cstdint>
#include <vector>
#include <mutex>
//
// SDL Audio capture
//
class audio_async {
public:
audio_async(int len_ms);
~audio_async();
bool init(int capture_id, int sample_rate);
// start capturing audio via the provided SDL callback
// keep last len_ms seconds of audio in a circular buffer
bool resume();
bool pause();
bool clear();
// callback to be called by SDL
void callback(uint8_t * stream, int len);
// get audio data from the circular buffer
void get(int ms, std::vector<float> & audio);
private:
SDL_AudioDeviceID m_dev_id_in = 0;
int m_len_ms = 0;
int m_sample_rate = 0;
std::atomic_bool m_running;
std::mutex m_mutex;
std::vector<float> m_audio;
std::vector<float> m_audio_new;
size_t m_audio_pos = 0;
size_t m_audio_len = 0;
};
// Return false if need to quit
bool sdl_poll_events();

@ -1,162 +0,0 @@
#include "common.h"
// third-party utilities
// use your favorite implementations
#define DR_WAV_IMPLEMENTATION
#include "dr_wav.h"
#include <cmath>
#include <regex>
#ifndef M_PI
#define M_PI 3.14159265358979323846
#endif
std::string trim(const std::string & s) {
std::regex e("^\\s+|\\s+$");
return std::regex_replace(s, e, "");
}
std::string replace(const std::string & s, const std::string & from, const std::string & to) {
std::string result = s;
size_t pos = 0;
while ((pos = result.find(from, pos)) != std::string::npos) {
result.replace(pos, from.length(), to);
pos += to.length();
}
return result;
}
bool read_wav(const std::string & fname, std::vector<float>& pcmf32, std::vector<std::vector<float>>& pcmf32s, bool stereo) {
drwav wav;
std::vector<uint8_t> wav_data; // used for pipe input from stdin
if (fname == "-") {
{
uint8_t buf[1024];
while (true)
{
const size_t n = fread(buf, 1, sizeof(buf), stdin);
if (n == 0) {
break;
}
wav_data.insert(wav_data.end(), buf, buf + n);
}
}
if (drwav_init_memory(&wav, wav_data.data(), wav_data.size(), nullptr) == false) {
fprintf(stderr, "error: failed to open WAV file from stdin\n");
return false;
}
fprintf(stderr, "%s: read %zu bytes from stdin\n", __func__, wav_data.size());
}
else if (drwav_init_file(&wav, fname.c_str(), nullptr) == false) {
fprintf(stderr, "error: failed to open '%s' as WAV file\n", fname.c_str());
return false;
}
if (wav.channels != 1 && wav.channels != 2) {
fprintf(stderr, "%s: WAV file '%s' must be mono or stereo\n", __func__, fname.c_str());
return false;
}
if (stereo && wav.channels != 2) {
fprintf(stderr, "%s: WAV file '%s' must be stereo for diarization\n", __func__, fname.c_str());
return false;
}
if (wav.sampleRate != COMMON_SAMPLE_RATE) {
fprintf(stderr, "%s: WAV file '%s' must be %i kHz\n", __func__, fname.c_str(), COMMON_SAMPLE_RATE/1000);
return false;
}
if (wav.bitsPerSample != 16) {
fprintf(stderr, "%s: WAV file '%s' must be 16-bit\n", __func__, fname.c_str());
return false;
}
const uint64_t n = wav_data.empty() ? wav.totalPCMFrameCount : wav_data.size()/(wav.channels*wav.bitsPerSample/8);
std::vector<int16_t> pcm16;
pcm16.resize(n*wav.channels);
drwav_read_pcm_frames_s16(&wav, n, pcm16.data());
drwav_uninit(&wav);
// convert to mono, float
pcmf32.resize(n);
if (wav.channels == 1) {
for (uint64_t i = 0; i < n; i++) {
pcmf32[i] = float(pcm16[i])/32768.0f;
}
} else {
for (uint64_t i = 0; i < n; i++) {
pcmf32[i] = float(pcm16[2*i] + pcm16[2*i + 1])/65536.0f;
}
}
if (stereo) {
// convert to stereo, float
pcmf32s.resize(2);
pcmf32s[0].resize(n);
pcmf32s[1].resize(n);
for (uint64_t i = 0; i < n; i++) {
pcmf32s[0][i] = float(pcm16[2*i])/32768.0f;
pcmf32s[1][i] = float(pcm16[2*i + 1])/32768.0f;
}
}
return true;
}
void high_pass_filter(std::vector<float> & data, float cutoff, float sample_rate) {
const float rc = 1.0f / (2.0f * M_PI * cutoff);
const float dt = 1.0f / sample_rate;
const float alpha = dt / (rc + dt);
float y = data[0];
for (size_t i = 1; i < data.size(); i++) {
y = alpha * (y + data[i] - data[i - 1]);
data[i] = y;
}
}
bool vad_simple(std::vector<float> & pcmf32, int sample_rate, int last_ms, float vad_thold, float freq_thold, bool verbose) {
const int n_samples = pcmf32.size();
const int n_samples_last = (sample_rate * last_ms) / 1000;
if (n_samples_last >= n_samples) {
// not enough samples - assume no speech
return false;
}
if (freq_thold > 0.0f) {
high_pass_filter(pcmf32, freq_thold, sample_rate);
}
float energy_all = 0.0f;
float energy_last = 0.0f;
for (int i = 0; i < n_samples; i++) {
energy_all += fabsf(pcmf32[i]);
if (i >= n_samples - n_samples_last) {
energy_last += fabsf(pcmf32[i]);
}
}
energy_all /= n_samples;
energy_last /= n_samples_last;
if (verbose) {
fprintf(stderr, "%s: energy_all: %f, energy_last: %f, vad_thold: %f, freq_thold: %f\n", __func__, energy_all, energy_last, vad_thold, freq_thold);
}
if (energy_last > vad_thold*energy_all) {
return false;
}
return true;
}

@ -1,40 +0,0 @@
#pragma once
// needs to match WHISPER_SAMPLE_RATE
#define COMMON_SAMPLE_RATE 16000
#include <vector>
#include <string>
std::string trim(const std::string & s);
std::string replace(
const std::string & s,
const std::string & from,
const std::string & to);
// Read WAV audio file and store the PCM data into pcmf32
// The sample rate of the audio must be equal to COMMON_SAMPLE_RATE
// If stereo flag is set and the audio has 2 channels, the pcmf32s will contain 2 channel PCM
bool read_wav(
const std::string & fname,
std::vector<float> & pcmf32,
std::vector<std::vector<float>> & pcmf32s,
bool stereo);
// Apply a high-pass frequency filter to PCM audio
// Suppresses frequencies below cutoff Hz
void high_pass_filter(
std::vector<float> & data,
float cutoff,
float sample_rate);
// Basic voice activity detection (VAD) using audio energy adaptive threshold
bool vad_simple(
std::vector<float> & pcmf32,
int sample_rate,
int last_ms,
float vad_thold,
float freq_thold,
bool verbose);

@ -8,7 +8,7 @@ function convertTypedArray(src, type) {
var printTextarea = (function() {
var element = document.getElementById('output');
if (element) element.value = ''; // clear browser cache
if (element) element.alue = ''; // clear browser cache
return function(text) {
if (arguments.length > 1) text = Array.prototype.slice.call(arguments).join(' ');
console.log(text);
@ -88,15 +88,11 @@ async function fetchRemote(url, cbProgress, cbPrint) {
// - check if the data is already in the IndexedDB
// - if not, fetch it from the remote URL and store it in the IndexedDB
function loadRemote(url, dst, size_mb, cbProgress, cbReady, cbCancel, cbPrint) {
if (!navigator.storage || !navigator.storage.estimate) {
cbPrint('loadRemote: navigator.storage.estimate() is not supported');
} else {
// query the storage quota and print it
navigator.storage.estimate().then(function (estimate) {
cbPrint('loadRemote: storage quota: ' + estimate.quota + ' bytes');
cbPrint('loadRemote: storage usage: ' + estimate.usage + ' bytes');
});
}
// query the storage quota and print it
navigator.storage.estimate().then(function (estimate) {
cbPrint('loadRemote: storage quota: ' + estimate.quota + ' bytes');
cbPrint('loadRemote: storage usage: ' + estimate.usage + ' bytes');
});
// check if the data is already in the IndexedDB
var rq = indexedDB.open(dbName, dbVersion);

@ -100,7 +100,7 @@ while [ $running -eq 1 ]; do
err=$(cat /tmp/whisper-live.err | wc -l)
done
./main -t 8 -m ./models/ggml-${model}.bin -f /tmp/whisper-live.wav --no-timestamps -otxt 2> /tmp/whispererr | tail -n 1
./main -t 8 -m ./models/ggml-base.en.bin -f /tmp/whisper-live.wav --no-timestamps -otxt 2> /tmp/whispererr | tail -n 1
while [ $SECONDS -lt $((($i+1)*$step_s)) ]; do
sleep 1

@ -3,4 +3,4 @@ add_executable(${TARGET} main.cpp)
include(DefaultTargetOptions)
target_link_libraries(${TARGET} PRIVATE common whisper ${CMAKE_THREAD_LIBS_INIT})
target_link_libraries(${TARGET} PRIVATE whisper ${CMAKE_THREAD_LIBS_INIT})

@ -9,35 +9,25 @@ It can be used as a reference for using the `whisper.cpp` library in other proje
usage: ./main [options] file0.wav file1.wav ...
options:
-h, --help [default] show this help message and exit
-t N, --threads N [4 ] number of threads to use during computation
-p N, --processors N [1 ] number of processors to use during computation
-ot N, --offset-t N [0 ] time offset in milliseconds
-on N, --offset-n N [0 ] segment index offset
-d N, --duration N [0 ] duration of audio to process in milliseconds
-mc N, --max-context N [-1 ] maximum number of text context tokens to store
-ml N, --max-len N [0 ] maximum segment length in characters
-bo N, --best-of N [5 ] number of best candidates to keep
-bs N, --beam-size N [-1 ] beam size for beam search
-wt N, --word-thold N [0.01 ] word timestamp probability threshold
-et N, --entropy-thold N [2.40 ] entropy threshold for decoder fail
-lpt N, --logprob-thold N [-1.00 ] log probability threshold for decoder fail
-su, --speed-up [false ] speed up audio by x2 (reduced accuracy)
-tr, --translate [false ] translate from source language to english
-di, --diarize [false ] stereo audio diarization
-nf, --no-fallback [false ] do not use temperature fallback while decoding
-otxt, --output-txt [false ] output result in a text file
-ovtt, --output-vtt [false ] output result in a vtt file
-osrt, --output-srt [false ] output result in a srt file
-owts, --output-words [false ] output script for generating karaoke video
-ocsv, --output-csv [false ] output result in a CSV file
-of FNAME, --output-file FNAME [ ] output file path (without file extension)
-ps, --print-special [false ] print special tokens
-pc, --print-colors [false ] print colors
-pp, --print-progress [false ] print progress
-nt, --no-timestamps [true ] do not print timestamps
-l LANG, --language LANG [en ] spoken language ('auto' for auto-detect)
--prompt PROMPT [ ] initial prompt
-m FNAME, --model FNAME [models/ggml-base.en.bin] model path
-f FNAME, --file FNAME [ ] input WAV file path
-h, --help [default] show this help message and exit
-t N, --threads N [4 ] number of threads to use during computation
-p N, --processors N [1 ] number of processors to use during computation
-ot N, --offset-t N [0 ] time offset in milliseconds
-on N, --offset-n N [0 ] segment index offset
-d N, --duration N [0 ] duration of audio to process in milliseconds
-mc N, --max-context N [-1 ] maximum number of text context tokens to store
-ml N, --max-len N [0 ] maximum segment length in characters
-wt N, --word-thold N [0.01 ] word timestamp probability threshold
-su, --speed-up [false ] speed up audio by x2 (reduced accuracy)
-tr, --translate [false ] translate from source language to english
-otxt, --output-txt [false ] output result in a text file
-ovtt, --output-vtt [false ] output result in a vtt file
-osrt, --output-srt [false ] output result in a srt file
-owts, --output-words [false ] output script for generating karaoke video
-ps, --print-special [false ] print special tokens
-pc, --print-colors [false ] print colors
-nt, --no-timestamps [true ] do not print timestamps
-l LANG, --language LANG [en ] spoken language
-m FNAME, --model FNAME [models/ggml-base.en.bin] model path
-f FNAME, --file FNAME [ ] input WAV file path
```

@ -1,7 +1,10 @@
#include "common.h"
#include "whisper.h"
// third-party utilities
// use your favorite implementations
#define DR_WAV_IMPLEMENTATION
#include "dr_wav.h"
#include <cmath>
#include <fstream>
#include <cstdio>
@ -50,24 +53,22 @@ void replace_all(std::string & s, const std::string & search, const std::string
// command-line parameters
struct whisper_params {
int32_t n_threads = std::min(4, (int32_t) std::thread::hardware_concurrency());
int32_t n_processors = 1;
int32_t offset_t_ms = 0;
int32_t offset_n = 0;
int32_t duration_ms = 0;
int32_t n_processors = 1;
int32_t offset_t_ms = 0;
int32_t offset_n = 0;
int32_t duration_ms = 0;
int32_t max_context = -1;
int32_t max_len = 0;
int32_t best_of = 5;
int32_t max_len = 0;
int32_t best_of = 5;
int32_t beam_size = -1;
float word_thold = 0.01f;
float entropy_thold = 2.40f;
float logprob_thold = -1.00f;
float word_thold = 0.01f;
float entropy_thold = 2.4f;
float logprob_thold = -1.0f;
bool speed_up = false;
bool translate = false;
bool diarize = false;
bool split_on_word = false;
bool no_fallback = false;
bool output_txt = false;
bool output_vtt = false;
bool output_srt = false;
@ -80,11 +81,9 @@ struct whisper_params {
std::string language = "en";
std::string prompt;
std::string font_path = "/System/Library/Fonts/Supplemental/Courier New Bold.ttf";
std::string model = "models/ggml-base.en.bin";
std::vector<std::string> fname_inp = {};
std::vector<std::string> fname_out = {};
};
void whisper_print_usage(int argc, char ** argv, const whisper_params & params);
@ -93,11 +92,6 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
for (int i = 1; i < argc; i++) {
std::string arg = argv[i];
if (arg == "-"){
params.fname_inp.push_back(arg);
continue;
}
if (arg[0] != '-') {
params.fname_inp.push_back(arg);
continue;
@ -122,15 +116,11 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
else if (arg == "-su" || arg == "--speed-up") { params.speed_up = true; }
else if (arg == "-tr" || arg == "--translate") { params.translate = true; }
else if (arg == "-di" || arg == "--diarize") { params.diarize = true; }
else if (arg == "-sow" || arg == "--split-on-word") { params.split_on_word = true; }
else if (arg == "-nf" || arg == "--no-fallback") { params.no_fallback = true; }
else if (arg == "-otxt" || arg == "--output-txt") { params.output_txt = true; }
else if (arg == "-ovtt" || arg == "--output-vtt") { params.output_vtt = true; }
else if (arg == "-osrt" || arg == "--output-srt") { params.output_srt = true; }
else if (arg == "-owts" || arg == "--output-words") { params.output_wts = true; }
else if (arg == "-fp" || arg == "--font-path") { params.font_path = argv[++i]; }
else if (arg == "-ocsv" || arg == "--output-csv") { params.output_csv = true; }
else if (arg == "-of" || arg == "--output-file") { params.fname_out.emplace_back(argv[++i]); }
else if (arg == "-ps" || arg == "--print-special") { params.print_special = true; }
else if (arg == "-pc" || arg == "--print-colors") { params.print_colors = true; }
else if (arg == "-pp" || arg == "--print-progress") { params.print_progress = true; }
@ -154,39 +144,35 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
fprintf(stderr, "usage: %s [options] file0.wav file1.wav ...\n", argv[0]);
fprintf(stderr, "\n");
fprintf(stderr, "options:\n");
fprintf(stderr, " -h, --help [default] show this help message and exit\n");
fprintf(stderr, " -t N, --threads N [%-7d] number of threads to use during computation\n", params.n_threads);
fprintf(stderr, " -p N, --processors N [%-7d] number of processors to use during computation\n", params.n_processors);
fprintf(stderr, " -ot N, --offset-t N [%-7d] time offset in milliseconds\n", params.offset_t_ms);
fprintf(stderr, " -on N, --offset-n N [%-7d] segment index offset\n", params.offset_n);
fprintf(stderr, " -d N, --duration N [%-7d] duration of audio to process in milliseconds\n", params.duration_ms);
fprintf(stderr, " -mc N, --max-context N [%-7d] maximum number of text context tokens to store\n", params.max_context);
fprintf(stderr, " -ml N, --max-len N [%-7d] maximum segment length in characters\n", params.max_len);
fprintf(stderr, " -sow, --split-on-word [%-7s] split on word rather than on token\n", params.split_on_word ? "true" : "false");
fprintf(stderr, " -bo N, --best-of N [%-7d] number of best candidates to keep\n", params.best_of);
fprintf(stderr, " -bs N, --beam-size N [%-7d] beam size for beam search\n", params.beam_size);
fprintf(stderr, " -wt N, --word-thold N [%-7.2f] word timestamp probability threshold\n", params.word_thold);
fprintf(stderr, " -et N, --entropy-thold N [%-7.2f] entropy threshold for decoder fail\n", params.entropy_thold);
fprintf(stderr, " -lpt N, --logprob-thold N [%-7.2f] log probability threshold for decoder fail\n", params.logprob_thold);
fprintf(stderr, " -su, --speed-up [%-7s] speed up audio by x2 (reduced accuracy)\n", params.speed_up ? "true" : "false");
fprintf(stderr, " -tr, --translate [%-7s] translate from source language to english\n", params.translate ? "true" : "false");
fprintf(stderr, " -di, --diarize [%-7s] stereo audio diarization\n", params.diarize ? "true" : "false");
fprintf(stderr, " -nf, --no-fallback [%-7s] do not use temperature fallback while decoding\n", params.no_fallback ? "true" : "false");
fprintf(stderr, " -otxt, --output-txt [%-7s] output result in a text file\n", params.output_txt ? "true" : "false");
fprintf(stderr, " -ovtt, --output-vtt [%-7s] output result in a vtt file\n", params.output_vtt ? "true" : "false");
fprintf(stderr, " -osrt, --output-srt [%-7s] output result in a srt file\n", params.output_srt ? "true" : "false");
fprintf(stderr, " -owts, --output-words [%-7s] output script for generating karaoke video\n", params.output_wts ? "true" : "false");
fprintf(stderr, " -fp, --font-path [%-7s] path to a monospace font for karaoke video\n", params.font_path.c_str());
fprintf(stderr, " -ocsv, --output-csv [%-7s] output result in a CSV file\n", params.output_csv ? "true" : "false");
fprintf(stderr, " -of FNAME, --output-file FNAME [%-7s] output file path (without file extension)\n", "");
fprintf(stderr, " -ps, --print-special [%-7s] print special tokens\n", params.print_special ? "true" : "false");
fprintf(stderr, " -pc, --print-colors [%-7s] print colors\n", params.print_colors ? "true" : "false");
fprintf(stderr, " -pp, --print-progress [%-7s] print progress\n", params.print_progress ? "true" : "false");
fprintf(stderr, " -nt, --no-timestamps [%-7s] do not print timestamps\n", params.no_timestamps ? "false" : "true");
fprintf(stderr, " -l LANG, --language LANG [%-7s] spoken language ('auto' for auto-detect)\n", params.language.c_str());
fprintf(stderr, " --prompt PROMPT [%-7s] initial prompt\n", params.prompt.c_str());
fprintf(stderr, " -m FNAME, --model FNAME [%-7s] model path\n", params.model.c_str());
fprintf(stderr, " -f FNAME, --file FNAME [%-7s] input WAV file path\n", "");
fprintf(stderr, " -h, --help [default] show this help message and exit\n");
fprintf(stderr, " -t N, --threads N [%-7d] number of threads to use during computation\n", params.n_threads);
fprintf(stderr, " -p N, --processors N [%-7d] number of processors to use during computation\n", params.n_processors);
fprintf(stderr, " -ot N, --offset-t N [%-7d] time offset in milliseconds\n", params.offset_t_ms);
fprintf(stderr, " -on N, --offset-n N [%-7d] segment index offset\n", params.offset_n);
fprintf(stderr, " -d N, --duration N [%-7d] duration of audio to process in milliseconds\n", params.duration_ms);
fprintf(stderr, " -mc N, --max-context N [%-7d] maximum number of text context tokens to store\n", params.max_context);
fprintf(stderr, " -ml N, --max-len N [%-7d] maximum segment length in characters\n", params.max_len);
fprintf(stderr, " -bo N, --best-of N [%-7d] number of best candidates to keep\n", params.best_of);
fprintf(stderr, " -bs N, --beam-size N [%-7d] beam size for beam search\n", params.beam_size);
fprintf(stderr, " -wt N, --word-thold N [%-7.2f] word timestamp probability threshold\n", params.word_thold);
fprintf(stderr, " -et N, --entropy-thold N [%-7.2f] entropy threshold for decoder fail\n", params.entropy_thold);
fprintf(stderr, " -lpt N, --logprob-thold N [%-7.2f] log probability threshold for decoder fail\n", params.logprob_thold);
fprintf(stderr, " -su, --speed-up [%-7s] speed up audio by x2 (reduced accuracy)\n", params.speed_up ? "true" : "false");
fprintf(stderr, " -tr, --translate [%-7s] translate from source language to english\n", params.translate ? "true" : "false");
fprintf(stderr, " -di, --diarize [%-7s] stereo audio diarization\n", params.diarize ? "true" : "false");
fprintf(stderr, " -otxt, --output-txt [%-7s] output result in a text file\n", params.output_txt ? "true" : "false");
fprintf(stderr, " -ovtt, --output-vtt [%-7s] output result in a vtt file\n", params.output_vtt ? "true" : "false");
fprintf(stderr, " -osrt, --output-srt [%-7s] output result in a srt file\n", params.output_srt ? "true" : "false");
fprintf(stderr, " -owts, --output-words [%-7s] output script for generating karaoke video\n", params.output_wts ? "true" : "false");
fprintf(stderr, " -ocsv, --output-csv [%-7s] output result in a CSV file\n", params.output_csv ? "true" : "false");
fprintf(stderr, " -ps, --print-special [%-7s] print special tokens\n", params.print_special ? "true" : "false");
fprintf(stderr, " -pc, --print-colors [%-7s] print colors\n", params.print_colors ? "true" : "false");
fprintf(stderr, " -pp, --print-progress [%-7s] print progress\n", params.print_progress ? "true" : "false");
fprintf(stderr, " -nt, --no-timestamps [%-7s] do not print timestamps\n", params.no_timestamps ? "false" : "true");
fprintf(stderr, " -l LANG, --language LANG [%-7s] spoken language ('auto' for auto-detect)\n", params.language.c_str());
fprintf(stderr, " --prompt PROMPT [%-7s] initial prompt\n", params.prompt.c_str());
fprintf(stderr, " -m FNAME, --model FNAME [%-7s] model path\n", params.model.c_str());
fprintf(stderr, " -f FNAME, --file FNAME [%-7s] input WAV file path\n", "");
fprintf(stderr, "\n");
}
@ -196,7 +182,7 @@ struct whisper_print_user_data {
const std::vector<std::vector<float>> * pcmf32s;
};
void whisper_print_segment_callback(struct whisper_context * ctx, struct whisper_state * /*state*/, int n_new, void * user_data) {
void whisper_print_segment_callback(struct whisper_context * ctx, int n_new, void * user_data) {
const auto & params = *((whisper_print_user_data *) user_data)->params;
const auto & pcmf32s = *((whisper_print_user_data *) user_data)->pcmf32s;
@ -355,14 +341,16 @@ bool output_csv(struct whisper_context * ctx, const char * fname) {
fprintf(stderr, "%s: saving output to '%s'\n", __func__, fname);
const int n_segments = whisper_full_n_segments(ctx);
fout << "start,end,text\n";
for (int i = 0; i < n_segments; ++i) {
const char * text = whisper_full_get_segment_text(ctx, i);
if (text[0] == ' ') {
text = text + sizeof(char); //whisper_full_get_segment_text() returns a string with leading space, point to the next character.
}
const int64_t t0 = whisper_full_get_segment_t0(ctx, i);
const int64_t t1 = whisper_full_get_segment_t1(ctx, i);
//need to multiply times returned from whisper_full_get_segment_t{0,1}() by 10 to get milliseconds.
fout << 10 * t0 << "," << 10 * t1 << ",\"" << text << "\"\n";
fout << 10 * t0 << ", " << 10 * t1 << ", \"" << text << "\"\n";
}
return true;
@ -371,18 +359,13 @@ bool output_csv(struct whisper_context * ctx, const char * fname) {
// karaoke video generation
// outputs a bash script that uses ffmpeg to generate a video with the subtitles
// TODO: font parameter adjustments
bool output_wts(struct whisper_context * ctx, const char * fname, const char * fname_inp, const whisper_params & params, float t_sec) {
bool output_wts(struct whisper_context * ctx, const char * fname, const char * fname_inp, const whisper_params & /*params*/, float t_sec) {
std::ofstream fout(fname);
fprintf(stderr, "%s: saving output to '%s'\n", __func__, fname);
static const char * font = params.font_path.c_str();
std::ifstream fin(font);
if (!fin.is_open()) {
fprintf(stderr, "%s: font not found at '%s', please specify a monospace font with -fp\n", __func__, font);
return false;
}
// TODO: become parameter
static const char * font = "/System/Library/Fonts/Supplemental/Courier New Bold.ttf";
fout << "#!/bin/bash" << "\n";
fout << "\n";
@ -531,14 +514,90 @@ int main(int argc, char ** argv) {
for (int f = 0; f < (int) params.fname_inp.size(); ++f) {
const auto fname_inp = params.fname_inp[f];
const auto fname_out = f < (int) params.fname_out.size() && !params.fname_out[f].empty() ? params.fname_out[f] : params.fname_inp[f];
std::vector<float> pcmf32; // mono-channel F32 PCM
std::vector<float> pcmf32; // mono-channel F32 PCM
std::vector<std::vector<float>> pcmf32s; // stereo-channel F32 PCM
if (!::read_wav(fname_inp, pcmf32, pcmf32s, params.diarize)) {
fprintf(stderr, "error: failed to read WAV file '%s'\n", fname_inp.c_str());
continue;
// WAV input
{
drwav wav;
std::vector<uint8_t> wav_data; // used for pipe input from stdin
if (fname_inp == "-") {
{
uint8_t buf[1024];
while (true)
{
const size_t n = fread(buf, 1, sizeof(buf), stdin);
if (n == 0) {
break;
}
wav_data.insert(wav_data.end(), buf, buf + n);
}
}
if (drwav_init_memory(&wav, wav_data.data(), wav_data.size(), nullptr) == false) {
fprintf(stderr, "error: failed to open WAV file from stdin\n");
return 4;
}
fprintf(stderr, "%s: read %zu bytes from stdin\n", __func__, wav_data.size());
}
else if (drwav_init_file(&wav, fname_inp.c_str(), nullptr) == false) {
fprintf(stderr, "error: failed to open '%s' as WAV file\n", fname_inp.c_str());
return 5;
}
if (wav.channels != 1 && wav.channels != 2) {
fprintf(stderr, "%s: WAV file '%s' must be mono or stereo\n", argv[0], fname_inp.c_str());
return 6;
}
if (params.diarize && wav.channels != 2 && params.no_timestamps == false) {
fprintf(stderr, "%s: WAV file '%s' must be stereo for diarization and timestamps have to be enabled\n", argv[0], fname_inp.c_str());
return 6;
}
if (wav.sampleRate != WHISPER_SAMPLE_RATE) {
fprintf(stderr, "%s: WAV file '%s' must be %i kHz\n", argv[0], fname_inp.c_str(), WHISPER_SAMPLE_RATE/1000);
return 8;
}
if (wav.bitsPerSample != 16) {
fprintf(stderr, "%s: WAV file '%s' must be 16-bit\n", argv[0], fname_inp.c_str());
return 9;
}
const uint64_t n = wav_data.empty() ? wav.totalPCMFrameCount : wav_data.size()/(wav.channels*wav.bitsPerSample/8);
std::vector<int16_t> pcm16;
pcm16.resize(n*wav.channels);
drwav_read_pcm_frames_s16(&wav, n, pcm16.data());
drwav_uninit(&wav);
// convert to mono, float
pcmf32.resize(n);
if (wav.channels == 1) {
for (uint64_t i = 0; i < n; i++) {
pcmf32[i] = float(pcm16[i])/32768.0f;
}
} else {
for (uint64_t i = 0; i < n; i++) {
pcmf32[i] = float(pcm16[2*i] + pcm16[2*i + 1])/65536.0f;
}
}
if (params.diarize) {
// convert to stereo, float
pcmf32s.resize(2);
pcmf32s[0].resize(n);
pcmf32s[1].resize(n);
for (uint64_t i = 0; i < n; i++) {
pcmf32s[0][i] = float(pcm16[2*i])/32768.0f;
pcmf32s[1][i] = float(pcm16[2*i + 1])/32768.0f;
}
}
}
// print system information
@ -587,20 +646,18 @@ int main(int argc, char ** argv) {
wparams.token_timestamps = params.output_wts || params.max_len > 0;
wparams.thold_pt = params.word_thold;
wparams.entropy_thold = params.entropy_thold;
wparams.logprob_thold = params.logprob_thold;
wparams.max_len = params.output_wts && params.max_len == 0 ? 60 : params.max_len;
wparams.split_on_word = params.split_on_word;
wparams.speed_up = params.speed_up;
wparams.prompt_tokens = prompt_tokens.empty() ? nullptr : prompt_tokens.data();
wparams.prompt_n_tokens = prompt_tokens.empty() ? 0 : prompt_tokens.size();
wparams.greedy.best_of = params.best_of;
wparams.beam_search.beam_size = params.beam_size;
wparams.temperature_inc = -1;
wparams.temperature_inc = params.no_fallback ? 0.0f : wparams.temperature_inc;
wparams.entropy_thold = params.entropy_thold;
wparams.logprob_thold = params.logprob_thold;
wparams.prompt_tokens = prompt_tokens.empty() ? nullptr : prompt_tokens.data();
wparams.prompt_n_tokens = prompt_tokens.empty() ? 0 : prompt_tokens.size();
whisper_print_user_data user_data = { &params, &pcmf32s };
@ -616,7 +673,7 @@ int main(int argc, char ** argv) {
{
static bool is_aborted = false; // NOTE: this should be atomic to avoid data race
wparams.encoder_begin_callback = [](struct whisper_context * /*ctx*/, struct whisper_state * /*state*/, void * user_data) {
wparams.encoder_begin_callback = [](struct whisper_context * /*ctx*/, void * user_data) {
bool is_aborted = *(bool*)user_data;
return !is_aborted;
};
@ -635,33 +692,34 @@ int main(int argc, char ** argv) {
// output to text file
if (params.output_txt) {
const auto fname_txt = fname_out + ".txt";
const auto fname_txt = fname_inp + ".txt";
output_txt(ctx, fname_txt.c_str());
}
// output to VTT file
if (params.output_vtt) {
const auto fname_vtt = fname_out + ".vtt";
const auto fname_vtt = fname_inp + ".vtt";
output_vtt(ctx, fname_vtt.c_str());
}
// output to SRT file
if (params.output_srt) {
const auto fname_srt = fname_out + ".srt";
const auto fname_srt = fname_inp + ".srt";
output_srt(ctx, fname_srt.c_str(), params);
}
// output to WTS file
if (params.output_wts) {
const auto fname_wts = fname_out + ".wts";
const auto fname_wts = fname_inp + ".wts";
output_wts(ctx, fname_wts.c_str(), fname_inp.c_str(), params, float(pcmf32.size() + 1000)/WHISPER_SAMPLE_RATE);
}
// output to CSV file
// output to CSV file
if (params.output_csv) {
const auto fname_csv = fname_out + ".csv";
const auto fname_csv = fname_inp + ".csv";
output_csv(ctx, fname_csv.c_str());
}
}
}

@ -5,5 +5,6 @@ if (WHISPER_SUPPORT_SDL2)
include(DefaultTargetOptions)
target_link_libraries(${TARGET} PRIVATE common common-sdl whisper ${CMAKE_THREAD_LIBS_INIT})
target_include_directories(${TARGET} PRIVATE ${SDL2_INCLUDE_DIRS})
target_link_libraries(${TARGET} PRIVATE whisper ${SDL2_LIBRARIES} ${CMAKE_THREAD_LIBS_INIT})
endif ()

@ -3,16 +3,19 @@
// A very quick-n-dirty implementation serving mainly as a proof of concept.
//
#include "common.h"
#include "common-sdl.h"
#include "whisper.h"
#include <SDL.h>
#include <SDL_audio.h>
#include <atomic>
#include <cassert>
#include <cstdio>
#include <string>
#include <thread>
#include <vector>
#include <fstream>
#include <mutex>
// 500 -> 00:05.000
// 6000 -> 01:00.000
@ -113,6 +116,306 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
fprintf(stderr, "\n");
}
//
// SDL Audio capture
//
class audio_async {
public:
audio_async(int len_ms);
~audio_async();
bool init(int capture_id, int sample_rate);
// start capturing audio via the provided SDL callback
// keep last len_ms seconds of audio in a circular buffer
bool resume();
bool pause();
bool clear();
// callback to be called by SDL
void callback(uint8_t * stream, int len);
// get audio data from the circular buffer
void get(int ms, std::vector<float> & audio);
private:
SDL_AudioDeviceID m_dev_id_in = 0;
int m_len_ms = 0;
int m_sample_rate = 0;
std::atomic_bool m_running;
std::mutex m_mutex;
std::vector<float> m_audio;
std::vector<float> m_audio_new;
size_t m_audio_pos = 0;
size_t m_audio_len = 0;
};
audio_async::audio_async(int len_ms) {
m_len_ms = len_ms;
m_running = false;
}
audio_async::~audio_async() {
if (m_dev_id_in) {
SDL_CloseAudioDevice(m_dev_id_in);
}
}
bool audio_async::init(int capture_id, int sample_rate) {
SDL_LogSetPriority(SDL_LOG_CATEGORY_APPLICATION, SDL_LOG_PRIORITY_INFO);
if (SDL_Init(SDL_INIT_AUDIO) < 0) {
SDL_LogError(SDL_LOG_CATEGORY_APPLICATION, "Couldn't initialize SDL: %s\n", SDL_GetError());
return false;
}
SDL_SetHintWithPriority(SDL_HINT_AUDIO_RESAMPLING_MODE, "medium", SDL_HINT_OVERRIDE);
{
int nDevices = SDL_GetNumAudioDevices(SDL_TRUE);
fprintf(stderr, "%s: found %d capture devices:\n", __func__, nDevices);
for (int i = 0; i < nDevices; i++) {
fprintf(stderr, "%s: - Capture device #%d: '%s'\n", __func__, i, SDL_GetAudioDeviceName(i, SDL_TRUE));
}
}
SDL_AudioSpec capture_spec_requested;
SDL_AudioSpec capture_spec_obtained;
SDL_zero(capture_spec_requested);
SDL_zero(capture_spec_obtained);
capture_spec_requested.freq = sample_rate;
capture_spec_requested.format = AUDIO_F32;
capture_spec_requested.channels = 1;
capture_spec_requested.samples = 1024;
capture_spec_requested.callback = [](void * userdata, uint8_t * stream, int len) {
audio_async * audio = (audio_async *) userdata;
audio->callback(stream, len);
};
capture_spec_requested.userdata = this;
if (capture_id >= 0) {
fprintf(stderr, "%s: attempt to open capture device %d : '%s' ...\n", __func__, capture_id, SDL_GetAudioDeviceName(capture_id, SDL_TRUE));
m_dev_id_in = SDL_OpenAudioDevice(SDL_GetAudioDeviceName(capture_id, SDL_TRUE), SDL_TRUE, &capture_spec_requested, &capture_spec_obtained, 0);
} else {
fprintf(stderr, "%s: attempt to open default capture device ...\n", __func__);
m_dev_id_in = SDL_OpenAudioDevice(nullptr, SDL_TRUE, &capture_spec_requested, &capture_spec_obtained, 0);
}
if (!m_dev_id_in) {
fprintf(stderr, "%s: couldn't open an audio device for capture: %s!\n", __func__, SDL_GetError());
m_dev_id_in = 0;
return false;
} else {
fprintf(stderr, "%s: obtained spec for input device (SDL Id = %d):\n", __func__, m_dev_id_in);
fprintf(stderr, "%s: - sample rate: %d\n", __func__, capture_spec_obtained.freq);
fprintf(stderr, "%s: - format: %d (required: %d)\n", __func__, capture_spec_obtained.format,
capture_spec_requested.format);
fprintf(stderr, "%s: - channels: %d (required: %d)\n", __func__, capture_spec_obtained.channels,
capture_spec_requested.channels);
fprintf(stderr, "%s: - samples per frame: %d\n", __func__, capture_spec_obtained.samples);
}
m_sample_rate = capture_spec_obtained.freq;
m_audio.resize((m_sample_rate*m_len_ms)/1000);
return true;
}
bool audio_async::resume() {
if (!m_dev_id_in) {
fprintf(stderr, "%s: no audio device to resume!\n", __func__);
return false;
}
if (m_running) {
fprintf(stderr, "%s: already running!\n", __func__);
return false;
}
SDL_PauseAudioDevice(m_dev_id_in, 0);
m_running = true;
return true;
}
bool audio_async::pause() {
if (!m_dev_id_in) {
fprintf(stderr, "%s: no audio device to pause!\n", __func__);
return false;
}
if (!m_running) {
fprintf(stderr, "%s: already paused!\n", __func__);
return false;
}
SDL_PauseAudioDevice(m_dev_id_in, 1);
m_running = false;
return true;
}
bool audio_async::clear() {
if (!m_dev_id_in) {
fprintf(stderr, "%s: no audio device to clear!\n", __func__);
return false;
}
if (!m_running) {
fprintf(stderr, "%s: not running!\n", __func__);
return false;
}
{
std::lock_guard<std::mutex> lock(m_mutex);
m_audio_pos = 0;
m_audio_len = 0;
}
return true;
}
// callback to be called by SDL
void audio_async::callback(uint8_t * stream, int len) {
if (!m_running) {
return;
}
const size_t n_samples = len / sizeof(float);
m_audio_new.resize(n_samples);
memcpy(m_audio_new.data(), stream, n_samples * sizeof(float));
//fprintf(stderr, "%s: %zu samples, pos %zu, len %zu\n", __func__, n_samples, m_audio_pos, m_audio_len);
{
std::lock_guard<std::mutex> lock(m_mutex);
if (m_audio_pos + n_samples > m_audio.size()) {
const size_t n0 = m_audio.size() - m_audio_pos;
memcpy(&m_audio[m_audio_pos], stream, n0 * sizeof(float));
memcpy(&m_audio[0], &stream[n0], (n_samples - n0) * sizeof(float));
m_audio_pos = (m_audio_pos + n_samples) % m_audio.size();
m_audio_len = m_audio.size();
} else {
memcpy(&m_audio[m_audio_pos], stream, n_samples * sizeof(float));
m_audio_pos = (m_audio_pos + n_samples) % m_audio.size();
m_audio_len = std::min(m_audio_len + n_samples, m_audio.size());
}
}
}
void audio_async::get(int ms, std::vector<float> & result) {
if (!m_dev_id_in) {
fprintf(stderr, "%s: no audio device to get audio from!\n", __func__);
return;
}
if (!m_running) {
fprintf(stderr, "%s: not running!\n", __func__);
return;
}
result.clear();
{
std::lock_guard<std::mutex> lock(m_mutex);
if (ms <= 0) {
ms = m_len_ms;
}
size_t n_samples = (m_sample_rate * ms) / 1000;
if (n_samples > m_audio_len) {
n_samples = m_audio_len;
}
result.resize(n_samples);
int s0 = m_audio_pos - n_samples;
if (s0 < 0) {
s0 += m_audio.size();
}
if (s0 + n_samples > m_audio.size()) {
const size_t n0 = m_audio.size() - s0;
memcpy(result.data(), &m_audio[s0], n0 * sizeof(float));
memcpy(&result[n0], &m_audio[0], (n_samples - n0) * sizeof(float));
} else {
memcpy(result.data(), &m_audio[s0], n_samples * sizeof(float));
}
}
}
///////////////////////////
void high_pass_filter(std::vector<float> & data, float cutoff, float sample_rate) {
const float rc = 1.0f / (2.0f * M_PI * cutoff);
const float dt = 1.0f / sample_rate;
const float alpha = dt / (rc + dt);
float y = data[0];
for (size_t i = 1; i < data.size(); i++) {
y = alpha * (y + data[i] - data[i - 1]);
data[i] = y;
}
}
bool vad_simple(std::vector<float> & pcmf32, int sample_rate, int last_ms, float vad_thold, float freq_thold, bool verbose) {
const int n_samples = pcmf32.size();
const int n_samples_last = (sample_rate * last_ms) / 1000;
if (n_samples_last >= n_samples) {
// not enough samples - assume no speech
return false;
}
if (freq_thold > 0.0f) {
high_pass_filter(pcmf32, freq_thold, sample_rate);
}
float energy_all = 0.0f;
float energy_last = 0.0f;
for (int i = 0; i < n_samples; i++) {
energy_all += fabsf(pcmf32[i]);
if (i >= n_samples - n_samples_last) {
energy_last += fabsf(pcmf32[i]);
}
}
energy_all /= n_samples;
energy_last /= n_samples_last;
if (verbose) {
fprintf(stderr, "%s: energy_all: %f, energy_last: %f, vad_thold: %f, freq_thold: %f\n", __func__, energy_all, energy_last, vad_thold, freq_thold);
}
if (energy_last > vad_thold*energy_all) {
return false;
}
return true;
}
int main(int argc, char ** argv) {
whisper_params params;
@ -120,17 +423,16 @@ int main(int argc, char ** argv) {
return 1;
}
params.keep_ms = std::min(params.keep_ms, params.step_ms);
params.length_ms = std::max(params.length_ms, params.step_ms);
params.keep_ms = std::min(params.keep_ms, params.step_ms); // cannot be more than step_ms
const int n_samples_step = (1e-3*params.step_ms )*WHISPER_SAMPLE_RATE;
const int n_samples_len = (1e-3*params.length_ms)*WHISPER_SAMPLE_RATE;
const int n_samples_keep = (1e-3*params.keep_ms )*WHISPER_SAMPLE_RATE;
const int n_samples_30s = (1e-3*30000.0 )*WHISPER_SAMPLE_RATE;
const int n_samples_step = (params.step_ms *1e-3)*WHISPER_SAMPLE_RATE;
const int n_samples_len = (params.length_ms*1e-3)*WHISPER_SAMPLE_RATE;
const int n_samples_keep = (params.keep_ms *1e-3)*WHISPER_SAMPLE_RATE;
const int n_samples_30s = (30000 *1e-3)*WHISPER_SAMPLE_RATE;
const bool use_vad = n_samples_step <= 0; // sliding window mode uses VAD
const int n_new_line = !use_vad ? std::max(1, params.length_ms / params.step_ms - 1) : 1; // number of steps to print new line
const int n_new_line = !use_vad ? params.length_ms / params.step_ms - 1 : 1; // number of steps to print new line
params.no_timestamps = !use_vad;
params.no_context |= use_vad;
@ -214,7 +516,23 @@ int main(int argc, char ** argv) {
// main audio loop
while (is_running) {
// handle Ctrl + C
is_running = sdl_poll_events();
{
SDL_Event event;
while (SDL_PollEvent(&event)) {
switch (event.type) {
case SDL_QUIT:
{
is_running = false;
} break;
default:
break;
}
}
if (!is_running) {
break;
}
}
if (!is_running) {
break;
@ -237,7 +555,7 @@ int main(int argc, char ** argv) {
break;
}
std::this_thread::sleep_for(std::chrono::milliseconds(1));
SDL_Delay(1);
}
const int n_samples_new = pcmf32_new.size();
@ -268,7 +586,7 @@ int main(int argc, char ** argv) {
audio.get(2000, pcmf32_new);
if (::vad_simple(pcmf32_new, WHISPER_SAMPLE_RATE, 1000, params.vad_thold, params.freq_thold, false)) {
if (vad_simple(pcmf32_new, WHISPER_SAMPLE_RATE, 1000, params.vad_thold, params.freq_thold, false)) {
audio.get(params.length_ms, pcmf32);
} else {
std::this_thread::sleep_for(std::chrono::milliseconds(100));
@ -288,6 +606,7 @@ int main(int argc, char ** argv) {
wparams.print_realtime = false;
wparams.print_timestamps = !params.no_timestamps;
wparams.translate = params.translate;
wparams.no_context = true;
wparams.single_segment = !use_vad;
wparams.max_tokens = params.max_tokens;
wparams.language = params.language.c_str();

@ -7,7 +7,7 @@ if (WHISPER_SUPPORT_SDL2)
# TODO: this is temporary
# need to export ggml symbols for MSVC, but too lazy ..
add_executable(${TARGET} talk.cpp gpt-2.cpp ../common.cpp ../common-sdl.cpp ../../ggml.c ../../whisper.cpp)
add_executable(${TARGET} talk.cpp gpt-2.cpp ../../ggml.c ../../whisper.cpp)
include(DefaultTargetOptions)

@ -1,14 +1,16 @@
// Talk with AI
//
#include "common.h"
#include "common-sdl.h"
#include "whisper.h"
#include "gpt-2.h"
#include <SDL.h>
#include <SDL_audio.h>
#include <cassert>
#include <cstdio>
#include <fstream>
#include <mutex>
#include <regex>
#include <string>
#include <thread>
@ -103,6 +105,320 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
fprintf(stderr, "\n");
}
//
// SDL Audio capture
//
class audio_async {
public:
audio_async(int len_ms);
~audio_async();
bool init(int capture_id, int sample_rate);
// start capturing audio via the provided SDL callback
// keep last len_ms seconds of audio in a circular buffer
bool resume();
bool pause();
bool clear();
// callback to be called by SDL
void callback(uint8_t * stream, int len);
// get audio data from the circular buffer
void get(int ms, std::vector<float> & audio);
private:
SDL_AudioDeviceID m_dev_id_in = 0;
int m_len_ms = 0;
int m_sample_rate = 0;
bool m_running = false;
std::mutex m_mutex;
std::vector<float> m_audio;
std::vector<float> m_audio_new;
size_t m_audio_pos = 0;
size_t m_audio_len = 0;
};
audio_async::audio_async(int len_ms) {
m_len_ms = len_ms;
}
audio_async::~audio_async() {
if (m_dev_id_in) {
SDL_CloseAudioDevice(m_dev_id_in);
}
}
bool audio_async::init(int capture_id, int sample_rate) {
SDL_LogSetPriority(SDL_LOG_CATEGORY_APPLICATION, SDL_LOG_PRIORITY_INFO);
if (SDL_Init(SDL_INIT_AUDIO) < 0) {
SDL_LogError(SDL_LOG_CATEGORY_APPLICATION, "Couldn't initialize SDL: %s\n", SDL_GetError());
return false;
}
SDL_SetHintWithPriority(SDL_HINT_AUDIO_RESAMPLING_MODE, "medium", SDL_HINT_OVERRIDE);
{
int nDevices = SDL_GetNumAudioDevices(SDL_TRUE);
fprintf(stderr, "%s: found %d capture devices:\n", __func__, nDevices);
for (int i = 0; i < nDevices; i++) {
fprintf(stderr, "%s: - Capture device #%d: '%s'\n", __func__, i, SDL_GetAudioDeviceName(i, SDL_TRUE));
}
}
SDL_AudioSpec capture_spec_requested;
SDL_AudioSpec capture_spec_obtained;
SDL_zero(capture_spec_requested);
SDL_zero(capture_spec_obtained);
capture_spec_requested.freq = sample_rate;
capture_spec_requested.format = AUDIO_F32;
capture_spec_requested.channels = 1;
capture_spec_requested.samples = 1024;
capture_spec_requested.callback = [](void * userdata, uint8_t * stream, int len) {
audio_async * audio = (audio_async *) userdata;
audio->callback(stream, len);
};
capture_spec_requested.userdata = this;
if (capture_id >= 0) {
fprintf(stderr, "%s: attempt to open capture device %d : '%s' ...\n", __func__, capture_id, SDL_GetAudioDeviceName(capture_id, SDL_TRUE));
m_dev_id_in = SDL_OpenAudioDevice(SDL_GetAudioDeviceName(capture_id, SDL_TRUE), SDL_TRUE, &capture_spec_requested, &capture_spec_obtained, 0);
} else {
fprintf(stderr, "%s: attempt to open default capture device ...\n", __func__);
m_dev_id_in = SDL_OpenAudioDevice(nullptr, SDL_TRUE, &capture_spec_requested, &capture_spec_obtained, 0);
}
if (!m_dev_id_in) {
fprintf(stderr, "%s: couldn't open an audio device for capture: %s!\n", __func__, SDL_GetError());
m_dev_id_in = 0;
return false;
} else {
fprintf(stderr, "%s: obtained spec for input device (SDL Id = %d):\n", __func__, m_dev_id_in);
fprintf(stderr, "%s: - sample rate: %d\n", __func__, capture_spec_obtained.freq);
fprintf(stderr, "%s: - format: %d (required: %d)\n", __func__, capture_spec_obtained.format,
capture_spec_requested.format);
fprintf(stderr, "%s: - channels: %d (required: %d)\n", __func__, capture_spec_obtained.channels,
capture_spec_requested.channels);
fprintf(stderr, "%s: - samples per frame: %d\n", __func__, capture_spec_obtained.samples);
fprintf(stderr, "\n");
}
m_sample_rate = capture_spec_obtained.freq;
m_audio.resize((m_sample_rate*m_len_ms)/1000);
return true;
}
bool audio_async::resume() {
if (!m_dev_id_in) {
fprintf(stderr, "%s: no audio device to resume!\n", __func__);
return false;
}
if (m_running) {
fprintf(stderr, "%s: already running!\n", __func__);
return false;
}
SDL_PauseAudioDevice(m_dev_id_in, 0);
m_running = true;
return true;
}
bool audio_async::pause() {
if (!m_dev_id_in) {
fprintf(stderr, "%s: no audio device to pause!\n", __func__);
return false;
}
if (!m_running) {
fprintf(stderr, "%s: already paused!\n", __func__);
return false;
}
SDL_PauseAudioDevice(m_dev_id_in, 1);
m_running = false;
return true;
}
bool audio_async::clear() {
if (!m_dev_id_in) {
fprintf(stderr, "%s: no audio device to clear!\n", __func__);
return false;
}
if (!m_running) {
fprintf(stderr, "%s: not running!\n", __func__);
return false;
}
{
std::lock_guard<std::mutex> lock(m_mutex);
m_audio_pos = 0;
m_audio_len = 0;
}
return true;
}
// callback to be called by SDL
void audio_async::callback(uint8_t * stream, int len) {
if (!m_running) {
return;
}
const size_t n_samples = len / sizeof(float);
m_audio_new.resize(n_samples);
memcpy(m_audio_new.data(), stream, n_samples * sizeof(float));
//fprintf(stderr, "%s: %zu samples, pos %zu, len %zu\n", __func__, n_samples, m_audio_pos, m_audio_len);
{
std::lock_guard<std::mutex> lock(m_mutex);
if (m_audio_pos + n_samples > m_audio.size()) {
const size_t n0 = m_audio.size() - m_audio_pos;
memcpy(&m_audio[m_audio_pos], stream, n0 * sizeof(float));
memcpy(&m_audio[0], &stream[n0], (n_samples - n0) * sizeof(float));
m_audio_pos = (m_audio_pos + n_samples) % m_audio.size();
m_audio_len = m_audio.size();
} else {
memcpy(&m_audio[m_audio_pos], stream, n_samples * sizeof(float));
m_audio_pos = (m_audio_pos + n_samples) % m_audio.size();
m_audio_len = std::min(m_audio_len + n_samples, m_audio.size());
}
}
}
void audio_async::get(int ms, std::vector<float> & result) {
if (!m_dev_id_in) {
fprintf(stderr, "%s: no audio device to get audio from!\n", __func__);
return;
}
if (!m_running) {
fprintf(stderr, "%s: not running!\n", __func__);
return;
}
result.clear();
{
std::lock_guard<std::mutex> lock(m_mutex);
if (ms <= 0) {
ms = m_len_ms;
}
size_t n_samples = (m_sample_rate * ms) / 1000;
if (n_samples > m_audio_len) {
n_samples = m_audio_len;
}
result.resize(n_samples);
int s0 = m_audio_pos - n_samples;
if (s0 < 0) {
s0 += m_audio.size();
}
if (s0 + n_samples > m_audio.size()) {
const size_t n0 = m_audio.size() - s0;
memcpy(result.data(), &m_audio[s0], n0 * sizeof(float));
memcpy(&result[n0], &m_audio[0], (n_samples - n0) * sizeof(float));
} else {
memcpy(result.data(), &m_audio[s0], n_samples * sizeof(float));
}
}
}
///////////////////////////
std::string trim(const std::string & s) {
std::regex e("^\\s+|\\s+$");
return std::regex_replace(s, e, "");
}
std::string replace(const std::string & s, const std::string & from, const std::string & to) {
std::string result = s;
size_t pos = 0;
while ((pos = result.find(from, pos)) != std::string::npos) {
result.replace(pos, from.length(), to);
pos += to.length();
}
return result;
}
void high_pass_filter(std::vector<float> & data, float cutoff, float sample_rate) {
const float rc = 1.0f / (2.0f * M_PI * cutoff);
const float dt = 1.0f / sample_rate;
const float alpha = dt / (rc + dt);
float y = data[0];
for (size_t i = 1; i < data.size(); i++) {
y = alpha * (y + data[i] - data[i - 1]);
data[i] = y;
}
}
bool vad_simple(std::vector<float> & pcmf32, int sample_rate, int last_ms, float vad_thold, float freq_thold, bool verbose) {
const int n_samples = pcmf32.size();
const int n_samples_last = (sample_rate * last_ms) / 1000;
if (n_samples_last >= n_samples) {
// not enough samples - assume no speech
return false;
}
if (freq_thold > 0.0f) {
high_pass_filter(pcmf32, freq_thold, sample_rate);
}
float energy_all = 0.0f;
float energy_last = 0.0f;
for (int i = 0; i < n_samples; i++) {
energy_all += fabsf(pcmf32[i]);
if (i >= n_samples - n_samples_last) {
energy_last += fabsf(pcmf32[i]);
}
}
energy_all /= n_samples;
energy_last /= n_samples_last;
if (verbose) {
fprintf(stderr, "%s: energy_all: %f, energy_last: %f, vad_thold: %f, freq_thold: %f\n", __func__, energy_all, energy_last, vad_thold, freq_thold);
}
if (energy_last > vad_thold*energy_all) {
return false;
}
return true;
}
std::string transcribe(whisper_context * ctx, const whisper_params & params, const std::vector<float> & pcmf32, float & prob, int64_t & t_ms) {
const auto t_start = std::chrono::high_resolution_clock::now();
@ -241,10 +557,22 @@ int main(int argc, char ** argv) {
// main loop
while (is_running) {
// handle Ctrl + C
is_running = sdl_poll_events();
{
SDL_Event event;
while (SDL_PollEvent(&event)) {
switch (event.type) {
case SDL_QUIT:
{
is_running = false;
} break;
default:
break;
}
}
if (!is_running) {
break;
if (!is_running) {
break;
}
}
// delay
@ -255,7 +583,7 @@ int main(int argc, char ** argv) {
{
audio.get(2000, pcmf32_cur);
if (::vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1250, params.vad_thold, params.freq_thold, params.print_energy) || force_speak) {
if (vad_simple(pcmf32_cur, WHISPER_SAMPLE_RATE, 1250, params.vad_thold, params.freq_thold, params.print_energy) || force_speak) {
fprintf(stdout, "%s: Speech detected! Processing ...\n", __func__);
audio.get(params.voice_ms, pcmf32_cur);

@ -9,4 +9,4 @@ To use:
5. Select the "release" active build variant, and use Android Studio to run and deploy to your device.
[^1]: I recommend the tiny or base models for running on an Android device.
<img width="300" alt="image" src="https://user-images.githubusercontent.com/1670775/221613663-a17bf770-27ef-45ab-9a46-a5f99ba65d2a.jpg">
<img width="300" alt="image" src="https://user-images.githubusercontent.com/1991296/208154256-82d972dc-221b-48c4-bfcb-36ce68602f93.png">

@ -2,7 +2,6 @@ package com.whispercppdemo.ui.main
import androidx.compose.foundation.layout.*
import androidx.compose.foundation.rememberScrollState
import androidx.compose.foundation.text.selection.SelectionContainer
import androidx.compose.foundation.verticalScroll
import androidx.compose.material3.*
import androidx.compose.runtime.Composable
@ -20,7 +19,6 @@ fun MainScreen(viewModel: MainScreenViewModel) {
canTranscribe = viewModel.canTranscribe,
isRecording = viewModel.isRecording,
messageLog = viewModel.dataLog,
onBenchmarkTapped = viewModel::benchmark,
onTranscribeSampleTapped = viewModel::transcribeSample,
onRecordTapped = viewModel::toggleRecord
)
@ -32,7 +30,6 @@ private fun MainScreen(
canTranscribe: Boolean,
isRecording: Boolean,
messageLog: String,
onBenchmarkTapped: () -> Unit,
onTranscribeSampleTapped: () -> Unit,
onRecordTapped: () -> Unit
) {
@ -48,11 +45,8 @@ private fun MainScreen(
.padding(innerPadding)
.padding(16.dp)
) {
Column(verticalArrangement = Arrangement.SpaceBetween) {
Row(horizontalArrangement = Arrangement.SpaceBetween, modifier = Modifier.fillMaxWidth()) {
BenchmarkButton(enabled = canTranscribe, onClick = onBenchmarkTapped)
TranscribeSampleButton(enabled = canTranscribe, onClick = onTranscribeSampleTapped)
}
Row(horizontalArrangement = Arrangement.SpaceBetween) {
TranscribeSampleButton(enabled = canTranscribe, onClick = onTranscribeSampleTapped)
RecordButton(
enabled = canTranscribe,
isRecording = isRecording,
@ -66,16 +60,7 @@ private fun MainScreen(
@Composable
private fun MessageLog(log: String) {
SelectionContainer() {
Text(modifier = Modifier.verticalScroll(rememberScrollState()), text = log)
}
}
@Composable
private fun BenchmarkButton(enabled: Boolean, onClick: () -> Unit) {
Button(onClick = onClick, enabled = enabled) {
Text("Benchmark")
}
Text(modifier = Modifier.verticalScroll(rememberScrollState()), text = log)
}
@Composable

@ -41,15 +41,10 @@ class MainScreenViewModel(private val application: Application) : ViewModel() {
init {
viewModelScope.launch {
printSystemInfo()
loadData()
}
}
private suspend fun printSystemInfo() {
printMessage(String.format("System Info: %s\n", WhisperContext.getSystemInfo()));
}
private suspend fun loadData() {
printMessage("Loading data...\n")
try {
@ -86,29 +81,10 @@ class MainScreenViewModel(private val application: Application) : ViewModel() {
//whisperContext = WhisperContext.createContextFromFile(firstModel.absolutePath)
}
fun benchmark() = viewModelScope.launch {
runBenchmark(6)
}
fun transcribeSample() = viewModelScope.launch {
transcribeAudio(getFirstSample())
}
private suspend fun runBenchmark(nthreads: Int) {
if (!canTranscribe) {
return
}
canTranscribe = false
printMessage("Running benchmark. This will take minutes...\n")
whisperContext?.benchMemory(nthreads)?.let{ printMessage(it) }
printMessage("\n")
whisperContext?.benchGgmlMulMat(nthreads)?.let{ printMessage(it) }
canTranscribe = true
}
private suspend fun getFirstSample(): File = withContext(Dispatchers.IO) {
samplesPath.listFiles()!!.first()
}
@ -138,14 +114,11 @@ class MainScreenViewModel(private val application: Application) : ViewModel() {
canTranscribe = false
try {
printMessage("Reading wave samples... ")
printMessage("Reading wave samples...\n")
val data = readAudioSamples(file)
printMessage("${data.size / (16000 / 1000)} ms\n")
printMessage("Transcribing data...\n")
val start = System.currentTimeMillis()
val text = whisperContext?.transcribeData(data)
val elapsed = System.currentTimeMillis() - start
printMessage("Done ($elapsed ms): $text\n")
printMessage("Done: $text\n")
} catch (e: Exception) {
Log.w(LOG_TAG, e)
printMessage("${e.localizedMessage}\n")

@ -27,14 +27,6 @@ class WhisperContext private constructor(private var ptr: Long) {
}
}
suspend fun benchMemory(nthreads: Int): String = withContext(scope.coroutineContext) {
return@withContext WhisperLib.benchMemcpy(nthreads)
}
suspend fun benchGgmlMulMat(nthreads: Int): String = withContext(scope.coroutineContext) {
return@withContext WhisperLib.benchGgmlMulMat(nthreads)
}
suspend fun release() = withContext(scope.coroutineContext) {
if (ptr != 0L) {
WhisperLib.freeContext(ptr)
@ -74,10 +66,6 @@ class WhisperContext private constructor(private var ptr: Long) {
}
return WhisperContext(ptr)
}
fun getSystemInfo(): String {
return WhisperLib.getSystemInfo()
}
}
}
@ -86,7 +74,6 @@ private class WhisperLib {
init {
Log.d(LOG_TAG, "Primary ABI: ${Build.SUPPORTED_ABIS[0]}")
var loadVfpv4 = false
var loadV8fp16 = false
if (isArmEabiV7a()) {
// armeabi-v7a needs runtime detection support
val cpuInfo = cpuInfo()
@ -97,24 +84,11 @@ private class WhisperLib {
loadVfpv4 = true
}
}
} else if (isArmEabiV8a()) {
// ARMv8.2a needs runtime detection support
val cpuInfo = cpuInfo()
cpuInfo?.let {
Log.d(LOG_TAG, "CPU info: $cpuInfo")
if (cpuInfo.contains("fphp")) {
Log.d(LOG_TAG, "CPU supports fp16 arithmetic")
loadV8fp16 = true
}
}
}
if (loadVfpv4) {
Log.d(LOG_TAG, "Loading libwhisper_vfpv4.so")
System.loadLibrary("whisper_vfpv4")
} else if (loadV8fp16) {
Log.d(LOG_TAG, "Loading libwhisper_v8fp16_va.so")
System.loadLibrary("whisper_v8fp16_va")
} else {
Log.d(LOG_TAG, "Loading libwhisper.so")
System.loadLibrary("whisper")
@ -129,9 +103,6 @@ private class WhisperLib {
external fun fullTranscribe(contextPtr: Long, audioData: FloatArray)
external fun getTextSegmentCount(contextPtr: Long): Int
external fun getTextSegment(contextPtr: Long, index: Int): String
external fun getSystemInfo(): String
external fun benchMemcpy(nthread: Int): String
external fun benchGgmlMulMat(nthread: Int): String
}
}
@ -139,10 +110,6 @@ private fun isArmEabiV7a(): Boolean {
return Build.SUPPORTED_ABIS[0].equals("armeabi-v7a")
}
private fun isArmEabiV8a(): Boolean {
return Build.SUPPORTED_ABIS[0].equals("arm64-v8a")
}
private fun cpuInfo(): String? {
return try {
File("/proc/cpuinfo").inputStream().bufferedReader().use {

@ -12,15 +12,4 @@ ifeq ($(TARGET_ARCH_ABI),armeabi-v7a)
# https://android.googlesource.com/platform/ndk/+/master/sources/android/cpufeatures/cpu-features.h
LOCAL_CFLAGS += -mfpu=neon-vfpv4
include $(BUILD_SHARED_LIBRARY)
endif
ifeq ($(TARGET_ARCH_ABI),arm64-v8a)
include $(CLEAR_VARS)
LOCAL_MODULE := libwhisper_v8fp16_va
include $(LOCAL_PATH)/Whisper.mk
# Allow building NEON FMA code.
# https://android.googlesource.com/platform/ndk/+/master/sources/android/cpufeatures/cpu-features.h
LOCAL_CFLAGS += -march=armv8.2-a+fp16
include $(BUILD_SHARED_LIBRARY)
endif
endif

@ -6,7 +6,6 @@
#include <sys/sysinfo.h>
#include <string.h>
#include "whisper.h"
#include "ggml.h"
#define UNUSED(x) (void)(x)
#define TAG "JNI"
@ -214,30 +213,4 @@ Java_com_whispercppdemo_whisper_WhisperLib_00024Companion_getTextSegment(
const char *text = whisper_full_get_segment_text(context, index);
jstring string = (*env)->NewStringUTF(env, text);
return string;
}
JNIEXPORT jstring JNICALL
Java_com_whispercppdemo_whisper_WhisperLib_00024Companion_getSystemInfo(
JNIEnv *env, jobject thiz
) {
UNUSED(thiz);
const char *sysinfo = whisper_print_system_info();
jstring string = (*env)->NewStringUTF(env, sysinfo);
return string;
}
JNIEXPORT jstring JNICALL
Java_com_whispercppdemo_whisper_WhisperLib_00024Companion_benchMemcpy(JNIEnv *env, jobject thiz,
jint n_threads) {
UNUSED(thiz);
const char *bench_ggml_memcpy = whisper_bench_memcpy_str(n_threads);
jstring string = (*env)->NewStringUTF(env, bench_ggml_memcpy);
}
JNIEXPORT jstring JNICALL
Java_com_whispercppdemo_whisper_WhisperLib_00024Companion_benchGgmlMulMat(JNIEnv *env, jobject thiz,
jint n_threads) {
UNUSED(thiz);
const char *bench_ggml_mul_mat = whisper_bench_ggml_mul_mat_str(n_threads);
jstring string = (*env)->NewStringUTF(env, bench_ggml_mul_mat);
}
}

@ -32,8 +32,8 @@ set_target_properties(${TARGET} PROPERTIES LINK_FLAGS " \
--bind \
-s USE_PTHREADS=1 \
-s PTHREAD_POOL_SIZE=8 \
-s INITIAL_MEMORY=1500MB \
-s TOTAL_MEMORY=1500MB \
-s INITIAL_MEMORY=1024MB \
-s TOTAL_MEMORY=1024MB \
-s FORCE_FILESYSTEM=1 \
-s EXPORTED_RUNTIME_METHODS=\"['print', 'printErr', 'ccall', 'cwrap']\" \
${EXTRA_FLAGS} \

@ -46,12 +46,10 @@
<div id="model">
Whisper model: <span id="model-whisper-status"></span>
<button id="fetch-whisper-tiny-en" onclick="loadWhisper('tiny.en')">tiny.en (75 MB)</button>
<button id="fetch-whisper-tiny" onclick="loadWhisper('tiny')">tiny (75 MB)</button>
<button id="fetch-whisper-base-en" onclick="loadWhisper('base.en')">base.en (142 MB)</button>
<button id="fetch-whisper-base" onclick="loadWhisper('base')">base (142 MB)</button>
<button id="fetch-whisper-small-en" onclick="loadWhisper('small.en')">small.en (466 MB)</button>
<button id="fetch-whisper-small" onclick="loadWhisper('small')">small (466 MB)</button>
<button id="fetch-whisper-tiny-en" onclick="loadWhisper('tiny.en')">tiny.en (75 MB)</button>
<button id="fetch-whisper-tiny" onclick="loadWhisper('tiny')">tiny (75 MB)</button>
<button id="fetch-whisper-base-en" onclick="loadWhisper('base.en')">base.en (142 MB)</button>
<button id="fetch-whisper-base" onclick="loadWhisper('base')">base (142 MB)</button>
<span id="fetch-whisper-progress"></span>
<input type="file" id="whisper-file" name="file" onchange="loadFile(event, 'whisper.bin')" />
@ -62,8 +60,8 @@
<!-- radio button to select between file upload or microphone -->
<div id="input">
Input:
<input type="radio" id="file" name="input" value="file" checked="checked" onchange="changeInput('file')" /> <label for="file">File</label>
<input type="radio" id="mic" name="input" value="mic" onchange="changeInput('mic')" /> <label for="mic">Microphone</label>
<input type="radio" id="file" name="input" value="file" checked="checked" onchange="changeInput('file')" /> File
<input type="radio" id="mic" name="input" value="mic" onchange="changeInput('mic')" /> Microphone
</div>
<br>
@ -286,33 +284,27 @@
}
reader.readAsArrayBuffer(file);
document.getElementById('fetch-whisper-tiny-en' ).style.display = 'none';
document.getElementById('fetch-whisper-base-en' ).style.display = 'none';
document.getElementById('fetch-whisper-small-en').style.display = 'none';
document.getElementById('fetch-whisper-tiny' ).style.display = 'none';
document.getElementById('fetch-whisper-base' ).style.display = 'none';
document.getElementById('fetch-whisper-small' ).style.display = 'none';
document.getElementById('whisper-file' ).style.display = 'none';
document.getElementById('model-whisper-status' ).innerHTML = 'loaded model: ' + file.name;
document.getElementById('fetch-whisper-tiny-en').style.display = 'none';
document.getElementById('fetch-whisper-base-en').style.display = 'none';
document.getElementById('fetch-whisper-tiny' ).style.display = 'none';
document.getElementById('fetch-whisper-base' ).style.display = 'none';
document.getElementById('whisper-file' ).style.display = 'none';
document.getElementById('model-whisper-status' ).innerHTML = 'loaded model: ' + file.name;
}
function loadWhisper(model) {
let urls = {
'tiny.en': 'https://whisper.ggerganov.com/ggml-model-whisper-tiny.en.bin',
'tiny': 'https://whisper.ggerganov.com/ggml-model-whisper-tiny.bin',
'base.en': 'https://whisper.ggerganov.com/ggml-model-whisper-base.en.bin',
'base': 'https://whisper.ggerganov.com/ggml-model-whisper-base.bin',
'small.en': 'https://whisper.ggerganov.com/ggml-model-whisper-small.en.bin',
'small': 'https://whisper.ggerganov.com/ggml-model-whisper-small.bin',
'tiny.en': 'https://whisper.ggerganov.com/ggml-model-whisper-tiny.en.bin',
'tiny': 'https://whisper.ggerganov.com/ggml-model-whisper-tiny.bin',
'base.en': 'https://whisper.ggerganov.com/ggml-model-whisper-base.en.bin',
'base': 'https://whisper.ggerganov.com/ggml-model-whisper-base.bin',
};
let sizes = {
'tiny.en': 75,
'tiny': 75,
'base.en': 142,
'base': 142,
'small.en': 466,
'small': 466,
'tiny.en': 75,
'tiny': 75,
'base.en': 142,
'base': 142,
};
let url = urls[model];
@ -321,14 +313,12 @@
model_whisper = model;
document.getElementById('fetch-whisper-tiny-en' ).style.display = 'none';
document.getElementById('fetch-whisper-base-en' ).style.display = 'none';
document.getElementById('fetch-whisper-small-en').style.display = 'none';
document.getElementById('fetch-whisper-tiny' ).style.display = 'none';
document.getElementById('fetch-whisper-base' ).style.display = 'none';
document.getElementById('fetch-whisper-small' ).style.display = 'none';
document.getElementById('whisper-file' ).style.display = 'none';
document.getElementById('model-whisper-status' ).innerHTML = 'loading model: ' + model;
document.getElementById('fetch-whisper-tiny-en').style.display = 'none';
document.getElementById('fetch-whisper-base-en').style.display = 'none';
document.getElementById('fetch-whisper-tiny' ).style.display = 'none';
document.getElementById('fetch-whisper-base' ).style.display = 'none';
document.getElementById('whisper-file' ).style.display = 'none';
document.getElementById('model-whisper-status' ).innerHTML = 'loading model: ' + model;
cbProgress = function(p) {
let el = document.getElementById('fetch-whisper-progress');
@ -337,14 +327,12 @@
cbCancel = function() {
var el;
el = document.getElementById('fetch-whisper-tiny-en' ); if (el) el.style.display = 'inline-block';
el = document.getElementById('fetch-whisper-base-en' ); if (el) el.style.display = 'inline-block';
el = document.getElementById('fetch-whisper-small-en'); if (el) el.style.display = 'inline-block';
el = document.getElementById('fetch-whisper-tiny' ); if (el) el.style.display = 'inline-block';
el = document.getElementById('fetch-whisper-base' ); if (el) el.style.display = 'inline-block';
el = document.getElementById('fetch-whisper-small' ); if (el) el.style.display = 'inline-block';
el = document.getElementById('whisper-file' ); if (el) el.style.display = 'inline-block';
el = document.getElementById('model-whisper-status' ); if (el) el.innerHTML = '';
el = document.getElementById('fetch-whisper-tiny-en'); if (el) el.style.display = 'inline-block';
el = document.getElementById('fetch-whisper-base-en'); if (el) el.style.display = 'inline-block';
el = document.getElementById('fetch-whisper-tiny' ); if (el) el.style.display = 'inline-block';
el = document.getElementById('fetch-whisper-base' ); if (el) el.style.display = 'inline-block';
el = document.getElementById('whisper-file' ); if (el) el.style.display = 'inline-block';
el = document.getElementById('model-whisper-status' ); if (el) el.innerHTML = '';
};
loadRemote(url, dst, size_mb, cbProgress, storeFS, cbCancel, printTextarea);

@ -1,10 +1,20 @@
#!/usr/bin/env bash
# shellcheck disable=2086
# Small shell script to more easily automatically download and transcribe live stream VODs.
# This uses YT-DLP, ffmpeg and the CPP version of Whisper: https://github.com/ggerganov/whisper.cpp
# Use `./examples/yt-wsp.sh help` to print help info.
#
# Sample usage:
#
# git clone https://github.com/ggerganov/whisper.cpp
# cd whisper.cpp
# make
# ./examples/yt-wsp.sh https://www.youtube.com/watch?v=1234567890
#
# MIT License
# Copyright (c) 2022 Daniils Petrovs
# Copyright (c) 2023 Jennifer Capasso
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
@ -24,178 +34,114 @@
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
# Small shell script to more easily automatically download and transcribe live stream VODs.
# This uses YT-DLP, ffmpeg and the CPP version of Whisper: https://github.com/ggerganov/whisper.cpp
# Use `./examples/yt-wsp.sh help` to print help info.
#
# Sample usage:
#
# git clone https://github.com/ggerganov/whisper.cpp
# cd whisper.cpp
# make
# ./examples/yt-wsp.sh https://www.youtube.com/watch?v=1234567890
#
set -Eeuo pipefail
# get script file location
SCRIPT_PATH="$(realpath -e ${BASH_SOURCE[0]})";
SCRIPT_DIR="${SCRIPT_PATH%/*}"
################################################################################
# Documentation on downloading models can be found in the whisper.cpp repo:
# https://github.com/ggerganov/whisper.cpp/#usage
#
# note: unless a multilingual model is specified, WHISPER_LANG will be ignored
# and the video will be transcribed as if the audio were in the English language
################################################################################
MODEL_PATH="${MODEL_PATH:-${SCRIPT_DIR}/../models/ggml-base.en.bin}"
################################################################################
# Where to find the whisper.cpp executable. default to the examples directory
# which holds this script in source control
################################################################################
WHISPER_EXECUTABLE="${WHISPER_EXECUTABLE:-${SCRIPT_DIR}/../main}";
# Set to desired language to be translated into english
WHISPER_LANG="${WHISPER_LANG:-en}";
# Default to 4 threads (this was most performant on my 2020 M1 MBP)
WHISPER_THREAD_COUNT="${WHISPER_THREAD_COUNT:-4}";
# You can find how to download models in the OG repo: https://github.com/ggerganov/whisper.cpp/#usage
MODEL_PATH="${MODEL_PATH:-models/ggml-base.en.bin}" # Set to a multilingual model if you want to translate from foreign lang to en
WHISPER_EXECUTABLE="${WHISPER_EXECUTABLE:-whisper}" # Where to find the whisper.cpp executable
WHISPER_LANG="${WHISPER_LANG:-en}" # Set to desired lang to translate from
msg() {
echo >&2 -e "${1-}"
}
cleanup() {
local -r clean_me="${1}";
if [ -d "${clean_me}" ]; then
msg "Cleaning up...";
rm -rf "${clean_me}";
else
msg "'${clean_me}' does not appear to be a directory!";
exit 1;
fi;
msg "Cleaning up..."
rm -rf "${temp_dir}" "vod-resampled.wav" "vod-resampled.wav.srt"
}
print_help() {
echo "################################################################################"
echo "Usage: ./examples/yt-wsp.sh <video_url>"
echo "# See configurable env variables in the script; there are many!"
echo "# This script will produce an MP4 muxed file in the working directory; it will"
echo "# be named for the title and id of the video."
echo "# passing in https://youtu.be/VYJtb2YXae8 produces a file named";
echo "# 'Why_we_all_need_subtitles_now-VYJtb2YXae8-res.mp4'"
echo "# Requirements: ffmpeg yt-dlp whisper.cpp"
echo "################################################################################"
echo "See configurable env variables in the script"
echo "This will produce an MP4 muxed file called res.mp4 in the working directory"
echo "Requirements: ffmpeg yt-dlp whisper"
echo "Whisper needs to be built into the main binary with make, then you can rename it to something like 'whisper' and add it to your PATH for convenience."
echo "E.g. in the root of Whisper.cpp, run: 'make && cp ./main /usr/local/bin/whisper'"
}
check_requirements() {
if ! command -v ffmpeg &>/dev/null; then
echo "ffmpeg is required: https://ffmpeg.org";
echo "ffmpeg is required (https://ffmpeg.org)."
exit 1
fi;
fi
if ! command -v yt-dlp &>/dev/null; then
echo "yt-dlp is required: https://github.com/yt-dlp/yt-dlp";
exit 1;
fi;
if ! command -v "${WHISPER_EXECUTABLE}" &>/dev/null; then
echo "The C++ implementation of Whisper is required: https://github.com/ggerganov/whisper.cpp"
echo "Sample usage:";
echo "";
echo " git clone https://github.com/ggerganov/whisper.cpp";
echo " cd whisper.cpp";
echo " make";
echo " ./examples/yt-wsp.sh https://www.youtube.com/watch?v=1234567890";
echo "";
exit 1;
fi;
echo "yt-dlp is required (https://github.com/yt-dlp/yt-dlp)."
exit 1
fi
if ! command -v "$WHISPER_EXECUTABLE" &>/dev/null; then
WHISPER_EXECUTABLE="./main"
if ! command -v "$WHISPER_EXECUTABLE" &>/dev/null; then
echo "Whisper is required (https://github.com/ggerganov/whisper.cpp):"
echo "Sample usage:"
echo ""
echo " git clone https://github.com/ggerganov/whisper.cpp"
echo " cd whisper.cpp"
echo " make"
echo " ./examples/yt-wsp.sh https://www.youtube.com/watch?v=1234567890"
echo ""
exit 1
fi
fi
}
if [[ "${#}" -lt 1 ]]; then
print_help;
exit 1;
if [[ $# -lt 1 ]]; then
print_help
exit 1
fi
if [[ "${1##-*}" == "help" ]]; then
print_help;
exit 0;
if [[ "$1" == "help" ]]; then
print_help
exit 0
fi
check_requirements;
################################################################################
# create a temporary directory to work in
# set the temp_dir and temp_filename variables
################################################################################
temp_dir="$(mktemp -d ${SCRIPT_DIR}/tmp.XXXXXX)";
temp_filename="${temp_dir}/yt-dlp-filename";
################################################################################
# for now we only take one argument
# TODO: a for loop
################################################################################
source_url="${1}"
title_name="";
msg "Downloading VOD...";
################################################################################
# Download the video, put the dynamic output filename into a variable.
# Optionally add --cookies-from-browser BROWSER[+KEYRING][:PROFILE][::CONTAINER]
# for videos only available to logged-in users.
################################################################################
temp_dir="tmp"
source_url="$1"
check_requirements
msg "Downloading VOD..."
# Optionally add --cookies-from-browser BROWSER[+KEYRING][:PROFILE][::CONTAINER] for members only VODs
yt-dlp \
-f "bestvideo[ext=mp4]+bestaudio[ext=m4a]/best[ext=mp4]/best" \
-o "${temp_dir}/%(title)s-%(id)s.vod.mp4" \
--print-to-file "%(filename)s" "${temp_filename}" \
--no-simulate \
--no-write-auto-subs \
--restrict-filenames \
--embed-thumbnail \
--embed-chapters \
--xattrs \
"${source_url}";
title_name="$(xargs basename -s .vod.mp4 < ${temp_filename})";
"${source_url}" -o "${temp_dir}/vod.mp4"
msg "Extracting audio and resampling...";
msg "Extracting audio and resampling..."
ffmpeg -i "${temp_dir}/${title_name}.vod.mp4" \
ffmpeg -i "${temp_dir}/vod.mp4" \
-hide_banner \
-vn \
-loglevel error \
-ar 16000 \
-ac 1 \
-c:a pcm_s16le \
-y \
"${temp_dir}/${title_name}.vod-resampled.wav";
-c:a \
pcm_s16le -y "vod-resampled.wav"
msg "Transcribing to subtitle file...";
msg "Whisper specified at: '${WHISPER_EXECUTABLE}'";
msg "Transcribing to subtitle file..."
msg "Whisper specified at: ${WHISPER_EXECUTABLE}"
"${WHISPER_EXECUTABLE}" \
$WHISPER_EXECUTABLE \
-m "${MODEL_PATH}" \
-l "${WHISPER_LANG}" \
-f "${temp_dir}/${title_name}.vod-resampled.wav" \
-t "${WHISPER_THREAD_COUNT}" \
-f "vod-resampled.wav" \
-t 8 \
-osrt \
--translate;
--translate
msg "Embedding subtitle track...";
msg "Embedding subtitle track..."
ffmpeg -i "${temp_dir}/${title_name}.vod.mp4" \
ffmpeg -i "${temp_dir}/vod.mp4" \
-hide_banner \
-loglevel error \
-i "${temp_dir}/${title_name}.vod-resampled.wav.srt" \
-i "vod-resampled.wav.srt" \
-c copy \
-c:s mov_text \
-y "${title_name}-res.mp4";
-y res.mp4
#cleanup "${temp_dir}";
cleanup
msg "Done! Your finished file is ready: ${title_name}-res.mp4";
msg "Done! Your finished file is ready: res.mp4"

@ -19,7 +19,7 @@ printf "\n"
./bench -w 1 -t 1 2>&1
printf "\n"
printf "Running ggml_mul_mat benchmark with $n_threads threads\n"
printf "Running ggml_mul_mat benchmark with " $n_threads " threads\n"
printf "\n"
./bench -w 2 -t $n_threads 2>&1

@ -1,70 +0,0 @@
# Benchmark word-level timestamps for different models
#
# This script takes two arguments
# - an audio file
# - [optional] path to a font file
# I'm using "/usr/share/fonts/truetype/freefont/FreeMono.ttf" on Ubuntu
if [ -z "$1" ]; then
echo "Usage: $0 <audio file> [font file]"
exit 1
fi
#TODO: Make this a command line parameter
#models="base small large"
#models="tiny.en tiny base.en base small.en small medium.en medium large-v1 large"
models="tiny.en base.en small.en medium.en large"
DURATION=$(ffprobe -i $1 -show_entries format=duration -v quiet -of csv="p=0")
DURATION=$(printf "%.2f" $DURATION)
echo "Input file duration: ${DURATION}s"
for model in $models; do
echo "Running $model"
COMMAND="./main -m models/ggml-$model.bin -owts -f $1 -of $1.$model"
if [ ! -z "$2" ]; then
COMMAND="$COMMAND -fp $2"
fi
#TODO: Surface errors better
# TIMEFMT is for zsh, TIMEFORMAT is for bash
EXECTIME=$({ TIMEFMT="%E";TIMEFORMAT=%E; time $COMMAND >/dev/null 2>&1; } 2>&1)
# Slightly different formats between zsh and bash
if [ "${EXECTIME: -1}" == "s" ]; then
EXECTIME=${EXECTIME::-1}
fi
RATIO=$(echo "$DURATION / $EXECTIME" | bc -l)
RATIO=$(printf "%.2f" $RATIO)
echo "Execution time: ${EXECTIME}s (${RATIO}x realtime)"
# If the file already exists, delete it
if [ -f $1.mp4 ]; then
rm $1.mp4
fi
bash $1.$model.wts >/dev/null 2>&1
mv $1.mp4 $1.$model.mp4
ffmpeg -y -f lavfi -i color=c=black:s=1200x50:d=$DURATION -vf "drawtext=fontfile=$2:fontsize=36:x=10:y=(h-text_h)/2:text='ggml-$model - ${EXECTIME}s (${RATIO}x realtime)':fontcolor=lightgrey" $1.$model.info.mp4 >/dev/null 2>&1
done
COMMAND="ffmpeg -y"
for model in $models; do
COMMAND="$COMMAND -i $1.$model.info.mp4 -i $1.$model.mp4"
done
COMMAND="$COMMAND -filter_complex \""
COUNT=0
for model in $models; do
COMMAND="$COMMAND[${COUNT}:v][$(($COUNT+1)):v]"
COUNT=$((COUNT+2))
done
COMMAND="$COMMAND vstack=inputs=${COUNT}[v]\" -map \"[v]\" -map 1:a $1.all.mp4 >/dev/null 2>&1"
echo $COMMAND
# Run the command
eval $COMMAND

136
ggml.c

@ -339,12 +339,8 @@ int64_t ggml_cycles_per_ms(void) {
#if defined(__cpp_lib_hardware_interference_size)
#define CACHE_LINE_SIZE hardware_destructive_interference_size
#else
#if defined(__POWER9_VECTOR__)
#define CACHE_LINE_SIZE 128
#else
#define CACHE_LINE_SIZE 64
#endif
#endif
static const size_t CACHE_LINE_SIZE_F32 = CACHE_LINE_SIZE/sizeof(float);
@ -613,12 +609,9 @@ static const size_t CACHE_LINE_SIZE_F32 = CACHE_LINE_SIZE/sizeof(float);
#define GGML_F16_VEC_LOAD(p, i) (i & 0x1) ? \
vec_extract_fp32_from_shorth(vec_xl(0, p - GGML_F16_EPR)) : \
vec_extract_fp32_from_shortl(vec_xl(0, p))
#define GGML_ENDIAN_BYTE(i) ((unsigned char *)&(uint16_t){1})[i]
#define GGML_F16_VEC_STORE(p, r, i) \
if (i & 0x1) \
vec_xst(vec_pack_to_short_fp32(r[i - GGML_ENDIAN_BYTE(1)], \
r[i - GGML_ENDIAN_BYTE(0)]), \
0, p - GGML_F16_EPR)
#define GGML_F16_VEC_STORE(p, r, i) \
if (i & 0x1) \
vec_xst(vec_pack_to_short_fp32(r[i], r[i - 1]), 0, p - GGML_F16_EPR)
#elif defined(__wasm_simd128__)
@ -1258,7 +1251,7 @@ static const char * GGML_OP_SYMBOL[GGML_OP_COUNT] = {
//
struct ggml_object {
size_t offs;
size_t offset;
size_t size;
struct ggml_object * next;
@ -1284,9 +1277,6 @@ struct ggml_context {
struct ggml_object * objects_begin;
struct ggml_object * objects_end;
struct ggml_scratch scratch;
struct ggml_scratch scratch_save;
};
struct ggml_context_container {
@ -1349,7 +1339,7 @@ inline static void ggml_critical_section_end(void) {
void ggml_print_object(const struct ggml_object * obj) {
GGML_PRINT(" - ggml_object: offset = %zu, size = %zu, next = %p\n",
obj->offs, obj->size, (const void *) obj->next);
obj->offset, obj->size, (const void *) obj->next);
}
void ggml_print_objects(const struct ggml_context * ctx) {
@ -1545,14 +1535,12 @@ struct ggml_context * ggml_init(struct ggml_init_params params) {
}
*ctx = (struct ggml_context) {
/*.mem_size =*/ params.mem_size,
/*.mem_buffer =*/ params.mem_buffer ? params.mem_buffer : malloc(params.mem_size),
/*.mem_buffer_owned =*/ params.mem_buffer ? false : true,
/*.n_objects =*/ 0,
/*.objects_begin =*/ NULL,
/*.objects_end =*/ NULL,
/*.scratch =*/ { 0, 0, NULL, },
/*.scratch_save =*/ { 0, 0, NULL, },
.mem_size = params.mem_size,
.mem_buffer = params.mem_buffer ? params.mem_buffer : malloc(params.mem_size),
.mem_buffer_owned = params.mem_buffer ? false : true,
.n_objects = 0,
.objects_begin = NULL,
.objects_end = NULL,
};
ggml_assert_aligned(ctx->mem_buffer);
@ -1575,7 +1563,7 @@ void ggml_free(struct ggml_context * ctx) {
g_state.contexts[i].used = false;
GGML_PRINT_DEBUG("%s: context %d with %d objects has been freed. memory used = %zu\n",
__func__, i, ctx->n_objects, ctx->objects_end->offs + ctx->objects_end->size);
__func__, i, ctx->n_objects, ctx->objects_end->offset + ctx->objects_end->size);
if (ctx->mem_buffer_owned) {
free(ctx->mem_buffer);
@ -1594,15 +1582,7 @@ void ggml_free(struct ggml_context * ctx) {
}
size_t ggml_used_mem(const struct ggml_context * ctx) {
return ctx->objects_end->offs + ctx->objects_end->size;
}
size_t ggml_set_scratch(struct ggml_context * ctx, struct ggml_scratch scratch) {
const size_t result = ctx->scratch.data ? ctx->scratch.offs : 0;
ctx->scratch = scratch;
return result;
return ctx->objects_end->offset + ctx->objects_end->size;
}
////////////////////////////////////////////////////////////////////////////////
@ -1616,9 +1596,9 @@ struct ggml_tensor * ggml_new_tensor_impl(
// always insert objects at the end of the context's memory pool
struct ggml_object * obj_cur = ctx->objects_end;
const size_t cur_offs = obj_cur == NULL ? 0 : obj_cur->offs;
const size_t cur_size = obj_cur == NULL ? 0 : obj_cur->size;
const size_t cur_end = cur_offs + cur_size;
const size_t cur_offset = obj_cur == NULL ? 0 : obj_cur->offset;
const size_t cur_size = obj_cur == NULL ? 0 : obj_cur->size;
const size_t cur_end = cur_offset + cur_size;
size_t size_needed = 0;
@ -1629,52 +1609,25 @@ struct ggml_tensor * ggml_new_tensor_impl(
}
// align to GGML_MEM_ALIGN
size_needed = ((size_needed + GGML_MEM_ALIGN - 1)/GGML_MEM_ALIGN)*GGML_MEM_ALIGN;
}
char * const mem_buffer = ctx->mem_buffer;
struct ggml_object * const obj_new = (struct ggml_object *)(mem_buffer + cur_end);
if (ctx->scratch.data == NULL || data != NULL) {
size_needed += sizeof(struct ggml_tensor);
if (cur_end + size_needed + GGML_OBJECT_SIZE > ctx->mem_size) {
GGML_PRINT("%s: not enough space in the context's memory pool (needed %zu, available %zu)\n",
__func__, cur_end + size_needed + GGML_OBJECT_SIZE, ctx->mem_size);
assert(false);
return NULL;
}
*obj_new = (struct ggml_object) {
.offs = cur_end + GGML_OBJECT_SIZE,
.size = size_needed,
.next = NULL,
};
} else {
if (ctx->scratch.offs + size_needed > ctx->scratch.size) {
GGML_PRINT("%s: not enough space in the scratch memory\n", __func__);
assert(false);
return NULL;
}
if (cur_end + sizeof(struct ggml_tensor) + GGML_OBJECT_SIZE > ctx->mem_size) {
GGML_PRINT("%s: not enough space in the context's memory pool (needed %zu, available %zu)\n",
__func__, cur_end + sizeof(struct ggml_tensor) + GGML_OBJECT_SIZE, ctx->mem_size);
assert(false);
return NULL;
}
}
size_needed += sizeof(struct ggml_tensor);
data = (char * const) ctx->scratch.data + ctx->scratch.offs;
if (cur_end + size_needed + GGML_OBJECT_SIZE > ctx->mem_size) {
GGML_PRINT("%s: not enough space in the context's memory pool\n", __func__);
assert(false);
return NULL;
}
*obj_new = (struct ggml_object) {
.offs = cur_end + GGML_OBJECT_SIZE,
.size = sizeof(struct ggml_tensor),
.next = NULL,
};
char * const mem_buffer = ctx->mem_buffer;
//printf("scratch offs = %zu, size_needed = %zu\n", ctx->scratch.offs, size_needed);
struct ggml_object * const obj_new = (struct ggml_object *)(mem_buffer + cur_end);
ctx->scratch.offs += size_needed;
}
*obj_new = (struct ggml_object) {
.offset = cur_end + GGML_OBJECT_SIZE,
.size = size_needed,
.next = NULL,
};
if (obj_cur != NULL) {
obj_cur->next = obj_new;
@ -1685,9 +1638,9 @@ struct ggml_tensor * ggml_new_tensor_impl(
ctx->objects_end = obj_new;
//printf("%s: inserted new object at %zu, size = %zu\n", __func__, cur_end, obj_new->size);
//GGML_PRINT_DEBUG("%s: inserted new object at %zu\n", __func__, cur_end);
struct ggml_tensor * const result = (struct ggml_tensor *)(mem_buffer + obj_new->offs);
struct ggml_tensor * const result = (struct ggml_tensor *)(mem_buffer + obj_new->offset);
ggml_assert_aligned(result);
@ -1730,7 +1683,7 @@ struct ggml_tensor * ggml_new_tensor(
struct ggml_context * ctx,
enum ggml_type type,
int n_dims,
const int * ne) {
const int* ne) {
return ggml_new_tensor_impl(ctx, type, n_dims, ne, NULL);
}
@ -1772,26 +1725,16 @@ struct ggml_tensor * ggml_new_tensor_4d(
}
struct ggml_tensor * ggml_new_i32(struct ggml_context * ctx, int32_t value) {
ctx->scratch_save = ctx->scratch;
ctx->scratch.data = NULL;
struct ggml_tensor * result = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, 1);
ctx->scratch = ctx->scratch_save;
ggml_set_i32(result, value);
return result;
}
struct ggml_tensor * ggml_new_f32(struct ggml_context * ctx, float value) {
ctx->scratch_save = ctx->scratch;
ctx->scratch.data = NULL;
struct ggml_tensor * result = ggml_new_tensor_1d(ctx, GGML_TYPE_F32, 1);
ctx->scratch = ctx->scratch_save;
ggml_set_f32(result, value);
return result;
@ -2400,7 +2343,7 @@ struct ggml_tensor * ggml_repeat(
result->op = GGML_OP_REPEAT;
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
result->src0 = a;
result->src1 = b;
result->src1 = NULL;
return result;
}
@ -3016,7 +2959,9 @@ struct ggml_tensor * ggml_diag_mask_inf(
// TODO: when implement backward, fix this:
//struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
struct ggml_tensor * result = ggml_view_tensor(ctx, a);
struct ggml_tensor * b = ggml_new_i32(ctx, n_past);
struct ggml_tensor * b = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, 1);
((int32_t *) b->data)[0] = n_past;
result->op = GGML_OP_DIAG_MASK_INF;
result->grad = is_node ? ggml_dup_tensor(ctx, result) : NULL;
@ -4348,9 +4293,7 @@ static bool ggml_compute_forward_mul_mat_use_blas(
const int ne1 = dst->ne[1];
// TODO: find the optimal values for these
if (ggml_is_contiguous(src0) && ggml_is_contiguous(src1) && (
(ne0 >= 32 && ne1 >= 32 && ne10 >= 32)
)) {
if (ggml_is_contiguous(src0) && ggml_is_contiguous(src1) && ne0 >= 32 && ne1 >= 32 && ne10 >= 32) {
//printf("BLAS: %d %d %d\n", ne0, ne1, ne10);
return true;
}
@ -7339,9 +7282,6 @@ void ggml_graph_compute(struct ggml_context * ctx, struct ggml_cgraph * cgraph)
node->n_tasks = 1; // TODO: this actually is doing nothing
// the threads are still spinning
cur = sizeof(float)*(node->src0->ne[0]*node->src0->ne[1]);
//printf("src0: ne0 = %d, ne1 = %d, ne = %d\n", node->src0->ne[0], node->src0->ne[1], node->src0->ne[0]*node->src0->ne[1]);
//printf("src1: ne0 = %d, ne1 = %d, ne = %d\n", node->src1->ne[0], node->src1->ne[1], node->src1->ne[0]*node->src1->ne[1]);
//printf("cur = %zu\n", cur);
} else {
cur = sizeof(ggml_fp16_t)*ggml_nelements(node->src1);
}

@ -301,13 +301,6 @@ struct ggml_cgraph {
int64_t perf_time_us;
};
// scratch buffer
struct ggml_scratch {
size_t offs;
size_t size;
void * data;
};
struct ggml_init_params {
// memory pool
size_t mem_size; // bytes
@ -334,8 +327,6 @@ void ggml_free(struct ggml_context * ctx);
size_t ggml_used_mem(const struct ggml_context * ctx);
size_t ggml_set_scratch(struct ggml_context * ctx, struct ggml_scratch scratch);
struct ggml_tensor * ggml_new_tensor(
struct ggml_context * ctx,
enum ggml_type type,

File diff suppressed because it is too large Load Diff

@ -66,7 +66,6 @@ extern "C" {
//
struct whisper_context;
struct whisper_state;
typedef int whisper_token;
@ -102,20 +101,11 @@ extern "C" {
WHISPER_API struct whisper_context * whisper_init_from_buffer(void * buffer, size_t buffer_size);
WHISPER_API struct whisper_context * whisper_init(struct whisper_model_loader * loader);
// These are the same as the above, but the internal state of the context is not allocated automatically
// It is the responsibility of the caller to allocate the state using whisper_init_state() (#523)
WHISPER_API struct whisper_context * whisper_init_from_file_no_state(const char * path_model);
WHISPER_API struct whisper_context * whisper_init_from_buffer_no_state(void * buffer, size_t buffer_size);
WHISPER_API struct whisper_context * whisper_init_no_state(struct whisper_model_loader * loader);
WHISPER_API struct whisper_state * whisper_init_state(struct whisper_context * ctx);
// Frees all allocated memory
WHISPER_API void whisper_free (struct whisper_context * ctx);
WHISPER_API void whisper_free_state(struct whisper_state * state);
// Frees all memory allocated by the model.
WHISPER_API void whisper_free(struct whisper_context * ctx);
// Convert RAW PCM audio to log mel spectrogram.
// The resulting spectrogram is stored inside the default state of the provided whisper context.
// The resulting spectrogram is stored inside the provided whisper context.
// Returns 0 on success
WHISPER_API int whisper_pcm_to_mel(
struct whisper_context * ctx,
@ -123,30 +113,7 @@ extern "C" {
int n_samples,
int n_threads);
WHISPER_API int whisper_pcm_to_mel_with_state(
struct whisper_context * ctx,
struct whisper_state * state,
const float * samples,
int n_samples,
int n_threads);
// Convert RAW PCM audio to log mel spectrogram but applies a Phase Vocoder to speed up the audio x2.
// The resulting spectrogram is stored inside the default state of the provided whisper context.
// Returns 0 on success
WHISPER_API int whisper_pcm_to_mel_phase_vocoder(
struct whisper_context * ctx,
const float * samples,
int n_samples,
int n_threads);
WHISPER_API int whisper_pcm_to_mel_phase_vocoder_with_state(
struct whisper_context * ctx,
struct whisper_state * state,
const float * samples,
int n_samples,
int n_threads);
// This can be used to set a custom log mel spectrogram inside the default state of the provided whisper context.
// This can be used to set a custom log mel spectrogram inside the provided whisper context.
// Use this instead of whisper_pcm_to_mel() if you want to provide your own log mel spectrogram.
// n_mel must be 80
// Returns 0 on success
@ -156,14 +123,7 @@ extern "C" {
int n_len,
int n_mel);
WHISPER_API int whisper_set_mel_with_state(
struct whisper_context * ctx,
struct whisper_state * state,
const float * data,
int n_len,
int n_mel);
// Run the Whisper encoder on the log mel spectrogram stored inside the default state in the provided whisper context.
// Run the Whisper encoder on the log mel spectrogram stored inside the provided whisper context.
// Make sure to call whisper_pcm_to_mel() or whisper_set_mel() first.
// offset can be used to specify the offset of the first frame in the spectrogram.
// Returns 0 on success
@ -172,12 +132,6 @@ extern "C" {
int offset,
int n_threads);
WHISPER_API int whisper_encode_with_state(
struct whisper_context * ctx,
struct whisper_state * state,
int offset,
int n_threads);
// Run the Whisper decoder to obtain the logits and probabilities for the next token.
// Make sure to call whisper_encode() first.
// tokens + n_tokens is the provided context for the decoder.
@ -191,14 +145,6 @@ extern "C" {
int n_past,
int n_threads);
WHISPER_API int whisper_decode_with_state(
struct whisper_context * ctx,
struct whisper_state * state,
const whisper_token * tokens,
int n_tokens,
int n_past,
int n_threads);
// Convert the provided text into tokens.
// The tokens pointer must be large enough to hold the resulting tokens.
// Returns the number of tokens on success, no more than n_max_tokens
@ -234,26 +180,17 @@ extern "C" {
int n_threads,
float * lang_probs);
WHISPER_API int whisper_lang_auto_detect_with_state(
struct whisper_context * ctx,
struct whisper_state * state,
int offset_ms,
int n_threads,
float * lang_probs);
WHISPER_API int whisper_n_len (struct whisper_context * ctx); // mel length
WHISPER_API int whisper_n_len_from_state(struct whisper_state * state); // mel length
WHISPER_API int whisper_n_vocab (struct whisper_context * ctx);
WHISPER_API int whisper_n_text_ctx (struct whisper_context * ctx);
WHISPER_API int whisper_n_audio_ctx (struct whisper_context * ctx);
WHISPER_API int whisper_is_multilingual (struct whisper_context * ctx);
WHISPER_API int whisper_n_len (struct whisper_context * ctx); // mel length
WHISPER_API int whisper_n_vocab (struct whisper_context * ctx);
WHISPER_API int whisper_n_text_ctx (struct whisper_context * ctx);
WHISPER_API int whisper_n_audio_ctx (struct whisper_context * ctx);
WHISPER_API int whisper_is_multilingual(struct whisper_context * ctx);
// Token logits obtained from the last call to whisper_decode()
// The logits for the last token are stored in the last row
// Rows: n_tokens
// Cols: n_vocab
WHISPER_API float * whisper_get_logits (struct whisper_context * ctx);
WHISPER_API float * whisper_get_logits_from_state(struct whisper_state * state);
WHISPER_API float * whisper_get_logits(struct whisper_context * ctx);
// Token Id -> String. Uses the vocabulary in the provided context
WHISPER_API const char * whisper_token_to_str(struct whisper_context * ctx, whisper_token token);
@ -271,7 +208,7 @@ extern "C" {
WHISPER_API whisper_token whisper_token_translate (void);
WHISPER_API whisper_token whisper_token_transcribe(void);
// Performance information from the default state.
// Performance information
WHISPER_API void whisper_print_timings(struct whisper_context * ctx);
WHISPER_API void whisper_reset_timings(struct whisper_context * ctx);
@ -289,23 +226,12 @@ extern "C" {
// Text segment callback
// Called on every newly generated text segment
// Use the whisper_full_...() functions to obtain the text segments
typedef void (*whisper_new_segment_callback)(struct whisper_context * ctx, struct whisper_state * state, int n_new, void * user_data);
typedef void (*whisper_new_segment_callback)(struct whisper_context * ctx, int n_new, void * user_data);
// Encoder begin callback
// If not NULL, called before the encoder starts
// If it returns false, the computation is aborted
typedef bool (*whisper_encoder_begin_callback)(struct whisper_context * ctx, struct whisper_state * state, void * user_data);
// Logits filter callback
// Can be used to modify the logits before sampling
// If not NULL, called after applying temperature to logits
typedef void (*whisper_logits_filter_callback)(
struct whisper_context * ctx,
struct whisper_state * state,
const whisper_token_data * tokens,
int n_tokens,
float * logits,
void * user_data);
typedef bool (*whisper_encoder_begin_callback)(struct whisper_context * ctx, void * user_data);
// Parameters for the whisper_full() function
// If you chnage the order or add new parameters, make sure to update the default values in whisper.cpp:
@ -331,7 +257,6 @@ extern "C" {
float thold_pt; // timestamp token probability threshold (~0.01)
float thold_ptsum; // timestamp token sum probability threshold (~0.01)
int max_len; // max segment length in characters
bool split_on_word; // split on word rather than on token (when used with max_len)
int max_tokens; // max tokens per segment (0 = no limit)
// [EXPERIMENTAL] speed-up techniques
@ -349,7 +274,6 @@ extern "C" {
// common decoding parameters:
bool suppress_blank; // ref: https://github.com/openai/whisper/blob/f82bc59f5ea234d4b97fb2860842ed38519f7e65/whisper/decoding.py#L89
bool suppress_non_speech_tokens; // ref: https://github.com/openai/whisper/blob/7858aa9c08d98f75575035ecd6481f462d66ca27/whisper/tokenizer.py#L224-L253
float temperature; // initial decoding temperature, ref: https://ai.stackexchange.com/a/32478
float max_initial_ts; // ref: https://github.com/openai/whisper/blob/f82bc59f5ea234d4b97fb2860842ed38519f7e65/whisper/decoding.py#L97
@ -379,16 +303,11 @@ extern "C" {
// called each time before the encoder starts
whisper_encoder_begin_callback encoder_begin_callback;
void * encoder_begin_callback_user_data;
// called by each decoder to filter obtained logits
whisper_logits_filter_callback logits_filter_callback;
void * logits_filter_callback_user_data;
};
WHISPER_API struct whisper_full_params whisper_full_default_params(enum whisper_sampling_strategy strategy);
// Run the entire model: PCM -> log mel spectrogram -> encoder -> decoder -> text
// Not thread safe for same context
// Uses the specified decoding strategy to obtain the text.
WHISPER_API int whisper_full(
struct whisper_context * ctx,
@ -396,16 +315,7 @@ extern "C" {
const float * samples,
int n_samples);
WHISPER_API int whisper_full_with_state(
struct whisper_context * ctx,
struct whisper_state * state,
struct whisper_full_params params,
const float * samples,
int n_samples);
// Split the input audio in chunks and process each chunk separately using whisper_full_with_state()
// Result is stored in the default state of the context
// Not thread safe if executed in parallel on the same context.
// Split the input audio in chunks and process each chunk separately using whisper_full()
// It seems this approach can offer some speedup in some cases.
// However, the transcription accuracy can be worse at the beginning and end of each chunk.
WHISPER_API int whisper_full_parallel(
@ -415,56 +325,37 @@ extern "C" {
int n_samples,
int n_processors);
// Number of generated text segments
// Number of generated text segments.
// A segment can be a few words, a sentence, or even a paragraph.
WHISPER_API int whisper_full_n_segments (struct whisper_context * ctx);
WHISPER_API int whisper_full_n_segments_from_state(struct whisper_state * state);
// Language id associated with the context's default state
WHISPER_API int whisper_full_lang_id(struct whisper_context * ctx);
// Language id associated with the provided state
WHISPER_API int whisper_full_lang_id_from_state(struct whisper_state * state);
// Get the start and end time of the specified segment
WHISPER_API int64_t whisper_full_get_segment_t0 (struct whisper_context * ctx, int i_segment);
WHISPER_API int64_t whisper_full_get_segment_t0_from_state(struct whisper_state * state, int i_segment);
WHISPER_API int64_t whisper_full_get_segment_t1 (struct whisper_context * ctx, int i_segment);
WHISPER_API int64_t whisper_full_get_segment_t1_from_state(struct whisper_state * state, int i_segment);
WHISPER_API int whisper_full_n_segments(struct whisper_context * ctx);
// Get the text of the specified segment
WHISPER_API const char * whisper_full_get_segment_text (struct whisper_context * ctx, int i_segment);
WHISPER_API const char * whisper_full_get_segment_text_from_state(struct whisper_state * state, int i_segment);
// Get the start and end time of the specified segment.
WHISPER_API int64_t whisper_full_get_segment_t0(struct whisper_context * ctx, int i_segment);
WHISPER_API int64_t whisper_full_get_segment_t1(struct whisper_context * ctx, int i_segment);
// Get number of tokens in the specified segment
WHISPER_API int whisper_full_n_tokens (struct whisper_context * ctx, int i_segment);
WHISPER_API int whisper_full_n_tokens_from_state(struct whisper_state * state, int i_segment);
// Get the text of the specified segment.
WHISPER_API const char * whisper_full_get_segment_text(struct whisper_context * ctx, int i_segment);
// Get the token text of the specified token in the specified segment
WHISPER_API const char * whisper_full_get_token_text (struct whisper_context * ctx, int i_segment, int i_token);
WHISPER_API const char * whisper_full_get_token_text_from_state(struct whisper_context * ctx, struct whisper_state * state, int i_segment, int i_token);
// Get number of tokens in the specified segment.
WHISPER_API int whisper_full_n_tokens(struct whisper_context * ctx, int i_segment);
WHISPER_API whisper_token whisper_full_get_token_id (struct whisper_context * ctx, int i_segment, int i_token);
WHISPER_API whisper_token whisper_full_get_token_id_from_state(struct whisper_state * state, int i_segment, int i_token);
// Get the token text of the specified token in the specified segment.
WHISPER_API const char * whisper_full_get_token_text(struct whisper_context * ctx, int i_segment, int i_token);
WHISPER_API whisper_token whisper_full_get_token_id (struct whisper_context * ctx, int i_segment, int i_token);
// Get token data for the specified token in the specified segment
// Get token data for the specified token in the specified segment.
// This contains probabilities, timestamps, etc.
WHISPER_API whisper_token_data whisper_full_get_token_data (struct whisper_context * ctx, int i_segment, int i_token);
WHISPER_API whisper_token_data whisper_full_get_token_data_from_state(struct whisper_state * state, int i_segment, int i_token);
WHISPER_API whisper_token_data whisper_full_get_token_data(struct whisper_context * ctx, int i_segment, int i_token);
// Get the probability of the specified token in the specified segment
WHISPER_API float whisper_full_get_token_p (struct whisper_context * ctx, int i_segment, int i_token);
WHISPER_API float whisper_full_get_token_p_from_state(struct whisper_state * state, int i_segment, int i_token);
// Get the probability of the specified token in the specified segment.
WHISPER_API float whisper_full_get_token_p(struct whisper_context * ctx, int i_segment, int i_token);
////////////////////////////////////////////////////////////////////////////
// Temporary helpers needed for exposing ggml interface
WHISPER_API int whisper_bench_memcpy(int n_threads);
WHISPER_API const char * whisper_bench_memcpy_str(int n_threads);
WHISPER_API int whisper_bench_ggml_mul_mat(int n_threads);
WHISPER_API const char * whisper_bench_ggml_mul_mat_str(int n_threads);
#ifdef __cplusplus
}

Loading…
Cancel
Save