Update README.md and finalize the whisper.wasm example

pull/70/head
Georgi Gerganov 2 years ago
parent 491ecd7056
commit 6b45e37b2b

@ -124,6 +124,7 @@ else()
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /arch:AVX2 /D_CRT_SECURE_NO_WARNINGS=1")
else()
if (EMSCRIPTEN)
# we require support for WASM SIMD 128-bit
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -pthread -msimd128")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -pthread")
else()

@ -90,7 +90,7 @@ libwhisper.a: ggml.o whisper.o
ar rcs libwhisper.a ggml.o whisper.o
clean:
rm -f *.o main libwhisper.a
rm -f *.o main stream libwhisper.a
#
# Examples

@ -289,7 +289,7 @@ You can download the converted models using the [download-ggml-model.sh](downloa
https://ggml.ggerganov.com
For more details, see the conversion script [convert-pt-to-ggml.py](convert-pt-to-ggml.py) or the README in [models](models).
For more details, see the conversion script [models/convert-pt-to-ggml.py](models/convert-pt-to-ggml.py) or the README in [models](models).
## Bindings

@ -1,3 +1,27 @@
# whisper.wasm
Live demo: https://whisper.ggerganov.com
Inference of [OpenAI's Whisper ASR model](https://github.com/openai/whisper) inside the browser
This example uses a WebAssembly (WASM) port of the [whisper.cpp](https://github.com/ggerganov/whisper.cpp)
implementation of the transformer to run the inference inside a web page. The audio data does not leave your computer -
it is processed locally on your machine. The performance is not great but you should be able to achieve x2 or x3
real-time for the `tiny` and `base` models on a modern CPU and browser (i.e. transcribe a 60 seconds audio in about
~20-30 seconds).
This WASM port utilizes [WASM SIMD 128-bit intrinsics](https://emcc.zcopy.site/docs/porting/simd/) so you have to make
sure that [your browser supports them](https://webassembly.org/roadmap/).
The example is capable of running all models up to size `small` inclusive. Beyond that, the memory requirements and
performance are unsatisfactory. The implementation currently support only the `Greedy` sampling strategy. Both
transcription and translation are supported.
Since the model data is quite big (74MB for the `tiny` model) you need to manually load the model into the web-page.
The example supports both loading audio from a file and recording audio from the microphone. The maximum length of the
audio is limited to 120 seconds.
## Live demo
Link: https://whisper.ggerganov.com
![image](https://user-images.githubusercontent.com/1991296/197348344-1a7fead8-3dae-4922-8b06-df223a206603.png)

@ -162,7 +162,7 @@
</tr>
</table>
<br><br>
<br>
<!-- textarea with height filling the rest of the page -->
<textarea id="output" rows="20"></textarea>
@ -254,6 +254,10 @@
return new type(buffer);
}
//
// load model
//
function loadFile(event, fname) {
var file = event.target.files[0] || null;
if (file == null) {
@ -281,6 +285,10 @@
reader.readAsArrayBuffer(file);
}
//
// audio file
//
function loadAudio(event) {
if (!context) {
context = new AudioContext({sampleRate: 16000});
@ -327,7 +335,7 @@
}
//
// Microphone
// microphone
//
var mediaRecorder = null;

@ -3,6 +3,6 @@
models=( "tiny.en" "tiny" "base.en" "base" "small.en" "small" "medium.en" "medium" "large" )
for model in "${models[@]}"; do
python3 convert-pt-to-ggml.py ~/.cache/whisper/$model.pt ../whisper models/
python3 models/convert-pt-to-ggml.py ~/.cache/whisper/$model.pt ../whisper models/
mv -v models/ggml-model.bin models/ggml-$model.bin
done

Loading…
Cancel
Save