You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
whisper.cpp/README.md

300 lines
12 KiB

2 years ago
# whisper.cpp
2 years ago
[![Actions Status](https://github.com/ggerganov/whisper.cpp/workflows/CI/badge.svg)](https://github.com/ggerganov/whisper.cpp/actions)
[![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](https://opensource.org/licenses/MIT)
High-performance inference of [OpenAI's Whisper](https://github.com/openai/whisper) automatic speech recognition (ASR) model:
2 years ago
- Plain C/C++ implementation without dependencies
- Apple silicon first-class citizen - optimized via Arm Neon and Accelerate framework
- AVX intrinsics support for x86 architectures
2 years ago
- Mixed F16 / F32 precision
- Low memory usage (Flash Attention + Flash Forward)
2 years ago
- Zero memory allocations at runtime
2 years ago
- Runs on the CPU
2 years ago
- [C-style API](https://github.com/ggerganov/whisper.cpp/blob/master/whisper.h)
2 years ago
- Supported platforms: Linux, Mac OS (Intel and Arm), Windows (MSVC and MinGW), Raspberry Pi, Android
2 years ago
2 years ago
## Usage
2 years ago
To build the main program, run `make`. You can then transcribe a `.wav` file like this:
```bash
./main -f input.wav
```
Before running the program, make sure to download one of the ggml Whisper models. For example:
```bash
bash ./download-ggml-model.sh base.en
```
---
2 years ago
For a quick demo, simply run `make base.en`:
2 years ago
```java
2 years ago
$ make base.en
2 years ago
cc -O3 -std=c11 -Wall -Wextra -Wno-unused-parameter -Wno-unused-function -pthread -DGGML_USE_ACCELERATE -c ggml.c
c++ -O3 -std=c++11 -Wall -Wextra -Wno-unused-parameter -Wno-unused-function -pthread -c whisper.cpp
2 years ago
c++ -O3 -std=c++11 -Wall -Wextra -Wno-unused-parameter -Wno-unused-function -pthread main.cpp whisper.o ggml.o -o main -framework Accelerate
./main -h
usage: ./main [options] file0.wav file1.wav ...
options:
-h, --help show this help message and exit
-s SEED, --seed SEED RNG seed (default: -1)
-t N, --threads N number of threads to use during computation (default: 4)
-o N, --offset N offset in milliseconds (default: 0)
-v, --verbose verbose output
--translate translate from source language to english
-otxt, --output-txt output result in a text file
-ovtt, --output-vtt output result in a vtt file
-osrt, --output-srt output result in a srt file
-ps, --print_special print special tokens
-nt, --no_timestamps do not print timestamps
-l LANG, --language LANG spoken language (default: en)
-m FNAME, --model FNAME model path (default: models/ggml-base.en.bin)
-f FNAME, --file FNAME input WAV file path
bash ./download-ggml-model.sh base.en
Downloading ggml model base.en ...
2 years ago
models/ggml-base.en.bin 100%[=============================================>] 141.11M 3.13MB/s in 79s
Done! Model 'base.en' saved in 'models/ggml-base.en.bin'
You can now use it like this:
$ ./main -m models/ggml-base.en.bin -f samples/jfk.wav
2 years ago
===============================================
Running base.en on all samples in ./samples ...
===============================================
----------------------------------------------
[+] Running base.en on samples/jfk.wav ... (run 'ffplay samples/jfk.wav' to listen)
----------------------------------------------
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
whisper_model_load: n_audio_head = 8
whisper_model_load: n_audio_layer = 6
whisper_model_load: n_text_ctx = 448
whisper_model_load: n_text_state = 512
whisper_model_load: n_text_head = 8
whisper_model_load: n_text_layer = 6
whisper_model_load: n_mels = 80
whisper_model_load: f16 = 1
whisper_model_load: type = 2
2 years ago
whisper_model_load: mem_required = 505.00 MB
2 years ago
whisper_model_load: adding 1607 extra tokens
whisper_model_load: ggml ctx size = 163.43 MB
whisper_model_load: memory size = 22.83 MB
2 years ago
whisper_model_load: model size = 140.54 MB
main: processing 'samples/jfk.wav' (176000 samples, 11.0 sec), 4 threads, lang = en, task = transcribe, timestamps = 1 ...
2 years ago
[00:00.000 --> 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.
2 years ago
2 years ago
whisper_print_timings: load time = 87.21 ms
whisper_print_timings: mel time = 24.26 ms
whisper_print_timings: sample time = 3.87 ms
whisper_print_timings: encode time = 323.67 ms / 53.94 ms per layer
whisper_print_timings: decode time = 83.25 ms / 13.87 ms per layer
whisper_print_timings: total time = 522.66 ms
2 years ago
```
The command downloads the `base.en` model converted to custom `ggml` format and runs the inference on all `.wav` samples in the folder `samples`.
2 years ago
For detailed usage instructions, run: `./main -h`
Note that `whisper.cpp` currently runs only with 16-bit WAV files, so make sure to convert your input before running the tool.
For example, you can use `ffmpeg` like this:
```java
ffmpeg -i input.mp3 -ar 16000 -ac 1 -c:a pcm_s16le output.wav
```
## More audio samples
2 years ago
If you want some extra audio samples to play with, simply run:
```
make samples
```
This will download a few more audio files from Wikipedia and convert them to 16-bit WAV format via `ffmpeg`.
You can download and run the other models as follows:
2 years ago
```
make tiny.en
make tiny
2 years ago
make base.en
make base
2 years ago
make small.en
make small
2 years ago
make medium.en
make medium
make large
2 years ago
```
2 years ago
## Another example
2 years ago
Here is another example of transcribing a [3:24 min speech](https://upload.wikimedia.org/wikipedia/commons/1/1f/George_W_Bush_Columbia_FINAL.ogg)
2 years ago
in about half a minute on a MacBook M1 Pro, using `medium.en` model:
2 years ago
```java
$ ./main -m models/ggml-medium.en.bin -f samples/gb1.wav -t 8
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
whisper_model_load: n_audio_head = 16
whisper_model_load: n_audio_layer = 24
whisper_model_load: n_text_ctx = 448
whisper_model_load: n_text_state = 1024
whisper_model_load: n_text_head = 16
whisper_model_load: n_text_layer = 24
whisper_model_load: n_mels = 80
whisper_model_load: f16 = 1
whisper_model_load: type = 4
2 years ago
whisper_model_load: mem_required = 2610.00 MB
whisper_model_load: adding 1607 extra tokens
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
2 years ago
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.
2 years ago
[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.
2 years ago
[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.
2 years ago
[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.
2 years ago
[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.
2 years ago
[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
```
2 years ago
## Real-time audio input example
This is a naive example of performing real-time inference on audio from your microphone.
2 years ago
The `stream` tool samples the audio every half a second and runs the transcription continously.
More info is available in [issue #10](https://github.com/ggerganov/whisper.cpp/issues/10).
2 years ago
```java
./stream -m ./models/ggml-base.en.bin -t 8 --step 500 --length 5000
2 years ago
```
2 years ago
https://user-images.githubusercontent.com/1991296/194935793-76afede7-cfa8-48d8-a80f-28ba83be7d09.mp4
2 years ago
The `stream` tool depends on SDL2 library to capture audio from the microphone. You can build it like this:
```bash
# Install SDL2 on Linux
sudo apt-get install libsdl2-dev
# Install SDL2 on Mac OS
brew install sdl2
make stream
```
2 years ago
## Implementation details
2 years ago
- The core tensor operations are implemented in C ([ggml.h](ggml.h) / [ggml.c](ggml.c))
- The high-level C-style API is implemented in C++ ([whisper.h](whisper.h) / [whisper.cpp](whisper.cpp))
- Simple usage is demonstrated in [main.cpp](main.cpp)
- Sample real-time audio transcription from the microphone is demonstrated in [stream.cpp](stream.cpp)
2 years ago
The tensor operators are optimized heavily for Apple silicon CPUs. Depending on the computation size, Arm Neon SIMD
instrisics or CBLAS Accelerate framwork routines are used. The latter are especially effective for bigger sizes since
the framwork utilizes the special-purpose AMX coprocessor available in modern Apple products.
2 years ago
## Limitations
- Inference only
- No GPU support
2 years ago
- Very basic greedy sampling scheme - always pick up the token with highest probability.
This should be similar to the [GreedyDecoder](https://github.com/openai/whisper/blob/main/whisper/decoding.py#L249-L274)
from the original python implementation, so in order to make a fair comparison between the 2 implementations, make sure
to run the python code with the following parameters:
```
whisper --best_of 1 --beam_size 1 ...
```
In the future, `whisper.cpp` will support more sampling strategies.
2 years ago
## Memory usage
| Model | Disk | Mem |
| --- | --- | --- |
| tiny | 75 MB | ~280 MB |
| base | 142 MB | ~430 MB |
| small | 466 MB | ~1.0 GB |
| medium | 1.5 GB | ~2.6 GB |
| large | 2.9 GB | ~4.7 GB |
2 years ago
## ggml format
The original models are converted to a custom binary format. This allows to pack everything needed into a single file:
- model parameters
- mel filters
- vocabulary
- weights
2 years ago
You can download the converted models using the [download-ggml-model.sh](download-ggml-model.sh) script or from here:
https://ggml.ggerganov.com
2 years ago
For more details, see the conversion script [convert-pt-to-ggml.py](convert-pt-to-ggml.py) or the README in [models](models).
2 years ago
## Bindings
- [X] Rust: [tazz4843/whisper-rs](https://github.com/tazz4843/whisper-rs)
- [ ] Python:
- [ ] Obj-C:
- [ ] Java: