mirror of https://github.com/ggerganov/whisper.cpp.git synced 2024-12-31 19:21:46 +01:00

History

Jhen-Jie Hong 0463028bc2 whisper : add context param to disable gpu (#1293 ) * whisper : check state->ctx_metal not null * whisper : add whisper_context_params { use_gpu } * whisper : new API with params & deprecate old API * examples : use no-gpu param && whisper_init_from_file_with_params * whisper.objc : enable metal & disable on simulator * whisper.swiftui, metal : enable metal & support load default.metallib * whisper.android : use new API * bindings : use new API * addon.node : fix build & test * bindings : updata java binding * bindings : add missing whisper_context_default_params_by_ref WHISPER_API for java * metal : use SWIFTPM_MODULE_BUNDLE for GGML_SWIFT and reuse library load * metal : move bundle var into block * metal : use SWIFT_PACKAGE instead of GGML_SWIFT * style : minor updates --------- Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>		2023-11-06 11:04:24 +02:00
..
.gitignore	wip : initial WASM port	2022-10-22 18:54:01 +03:00
CMakeLists.txt	Node.js package (#260 )	2022-12-12 20:17:27 +02:00
emscripten.cpp	whisper : add context param to disable gpu (#1293 )	2023-11-06 11:04:24 +02:00
libwhisper.worker.js	ggml : sync (ggml-alloc, GPU, eps, etc.) (#1220 )	2023-09-05 13:54:40 +03:00
package-tmpl.json	Node.js package (#260 )	2022-12-12 20:17:27 +02:00
package.json	release : v1.4.2	2023-05-14 19:06:45 +03:00
README.md	Update README.md	2022-12-12 20:33:09 +02:00
whisper.js	ggml : sync (ggml-alloc, GPU, eps, etc.) (#1220 )	2023-09-05 13:54:40 +03:00

README.md

whisper.cpp

Node.js package for Whisper speech recognition

Package: https://www.npmjs.com/package/whisper.cpp

Details

The performance is comparable to when running whisper.cpp in the browser via WASM.

The API is currently very rudimentary: bindings/javascript/emscripten.cpp

For sample usage check tests/test-whisper.js

Package building + test

# load emscripten
source /path/to/emsdk/emsdk_env.sh

# clone repo
git clone https://github.com/ggerganov/whisper.cpp
cd whisper.cpp

# grab base.en model
./models/download-ggml-model.sh base.en

# prepare PCM sample for testing
ffmpeg -i samples/jfk.wav -f f32le -acodec pcm_f32le samples/jfk.pcmf32

# build
mkdir build-em && cd build-em
emcmake cmake .. && make -j

# run test
node --experimental-wasm-threads --experimental-wasm-simd ../tests/test-whisper.js

# publish npm package
make publish-npm

Sample run

$ node --experimental-wasm-threads --experimental-wasm-simd ../tests/test-whisper.js

whisper_model_load: loading model from 'whisper.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
whisper_model_load: adding 1607 extra tokens
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 = 8 / 10 | AVX = 0 | AVX2 = 0 | AVX512 = 0 | NEON = 0 | F16C = 0 | FP16_VA = 0 | WASM_SIMD = 1 | BLAS = 0 | 

operator(): processing 176000 samples, 11.0 sec, 8 threads, 1 processors, lang = en, task = transcribe ...

[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:     load time =   162.37 ms
whisper_print_timings:      mel time =   183.70 ms
whisper_print_timings:   sample time =     4.27 ms
whisper_print_timings:   encode time =  8582.63 ms / 1430.44 ms per layer
whisper_print_timings:   decode time =   436.16 ms / 72.69 ms per layer
whisper_print_timings:    total time =  9370.90 ms