This commit is contained in:
Kar
2025-06-17 15:53:01 +05:30
commit 4d20931ecc
411 changed files with 180695 additions and 0 deletions

View File

@@ -0,0 +1 @@
publish.log

View File

@@ -0,0 +1,41 @@
set(TARGET libwhisper)
add_executable(${TARGET}
emscripten.cpp
)
target_link_libraries(${TARGET} PRIVATE
whisper
)
unset(EXTRA_FLAGS)
if (WHISPER_WASM_SINGLE_FILE)
set(EXTRA_FLAGS "-s SINGLE_FILE=1")
message(STATUS "Embedding WASM inside whisper.js")
add_custom_command(
TARGET ${TARGET} POST_BUILD
COMMAND ${CMAKE_COMMAND} -E copy
${CMAKE_BINARY_DIR}/bin/libwhisper.js
${CMAKE_CURRENT_SOURCE_DIR}/whisper.js
)
add_custom_command(
TARGET ${TARGET} POST_BUILD
COMMAND ${CMAKE_COMMAND} -E copy
${CMAKE_BINARY_DIR}/bin/libwhisper.worker.js
${CMAKE_CURRENT_SOURCE_DIR}/libwhisper.worker.js
)
endif()
set_target_properties(${TARGET} PROPERTIES LINK_FLAGS " \
--bind \
-s MODULARIZE=1 \
-s EXPORT_NAME=\"'whisper_factory'\" \
-s FORCE_FILESYSTEM=1 \
-s USE_PTHREADS=1 \
-s PTHREAD_POOL_SIZE=8 \
-s ALLOW_MEMORY_GROWTH=1 \
${EXTRA_FLAGS} \
")

View File

@@ -0,0 +1,78 @@
# 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](/bindings/javascript/emscripten.cpp)
For sample usage check [tests/test-whisper.js](/tests/test-whisper.js)
## Package building + test
```bash
# 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
```java
$ 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
```

View File

@@ -0,0 +1,93 @@
//
// This is the Javascript API of whisper.cpp
//
// Very crude at the moment.
// Feel free to contribute and make this better!
//
// See the tests/test-whisper.js for sample usage
//
#include "whisper.h"
#include <emscripten.h>
#include <emscripten/bind.h>
#include <thread>
#include <vector>
struct whisper_context * g_context;
EMSCRIPTEN_BINDINGS(whisper) {
emscripten::function("init", emscripten::optional_override([](const std::string & path_model) {
if (g_context == nullptr) {
g_context = whisper_init_from_file_with_params(path_model.c_str(), whisper_context_default_params());
if (g_context != nullptr) {
return true;
} else {
return false;
}
}
return false;
}));
emscripten::function("free", emscripten::optional_override([]() {
if (g_context) {
whisper_free(g_context);
g_context = nullptr;
}
}));
emscripten::function("full_default", emscripten::optional_override([](const emscripten::val & audio, const std::string & lang, bool translate) {
if (g_context == nullptr) {
return -1;
}
struct whisper_full_params params = whisper_full_default_params(whisper_sampling_strategy::WHISPER_SAMPLING_GREEDY);
params.print_realtime = true;
params.print_progress = false;
params.print_timestamps = true;
params.print_special = false;
params.translate = translate;
params.language = whisper_is_multilingual(g_context) ? lang.c_str() : "en";
params.n_threads = std::min(8, (int) std::thread::hardware_concurrency());
params.offset_ms = 0;
std::vector<float> pcmf32;
const int n = audio["length"].as<int>();
emscripten::val heap = emscripten::val::module_property("HEAPU8");
emscripten::val memory = heap["buffer"];
pcmf32.resize(n);
emscripten::val memoryView = audio["constructor"].new_(memory, reinterpret_cast<uintptr_t>(pcmf32.data()), n);
memoryView.call<void>("set", audio);
// print system information
{
printf("\n");
printf("system_info: n_threads = %d / %d | %s\n",
params.n_threads, std::thread::hardware_concurrency(), whisper_print_system_info());
printf("\n");
printf("%s: processing %d samples, %.1f sec, %d threads, %d processors, lang = %s, task = %s ...\n",
__func__, int(pcmf32.size()), float(pcmf32.size())/WHISPER_SAMPLE_RATE,
params.n_threads, 1,
params.language,
params.translate ? "translate" : "transcribe");
printf("\n");
}
// run whisper
{
whisper_reset_timings(g_context);
whisper_full(g_context, params, pcmf32.data(), pcmf32.size());
whisper_print_timings(g_context);
}
return 0;
}));
}

View File

@@ -0,0 +1 @@
"use strict";var Module={};var ENVIRONMENT_IS_NODE=typeof process=="object"&&typeof process.versions=="object"&&typeof process.versions.node=="string";if(ENVIRONMENT_IS_NODE){var nodeWorkerThreads=require("worker_threads");var parentPort=nodeWorkerThreads.parentPort;parentPort.on("message",data=>onmessage({data:data}));var fs=require("fs");Object.assign(global,{self:global,require:require,Module:Module,location:{href:__filename},Worker:nodeWorkerThreads.Worker,importScripts:f=>(0,eval)(fs.readFileSync(f,"utf8")+"//# sourceURL="+f),postMessage:msg=>parentPort.postMessage(msg),performance:global.performance||{now:Date.now}})}var initializedJS=false;function threadPrintErr(){var text=Array.prototype.slice.call(arguments).join(" ");if(ENVIRONMENT_IS_NODE){fs.writeSync(2,text+"\n");return}console.error(text)}function threadAlert(){var text=Array.prototype.slice.call(arguments).join(" ");postMessage({cmd:"alert",text:text,threadId:Module["_pthread_self"]()})}var err=threadPrintErr;self.alert=threadAlert;Module["instantiateWasm"]=(info,receiveInstance)=>{var module=Module["wasmModule"];Module["wasmModule"]=null;var instance=new WebAssembly.Instance(module,info);return receiveInstance(instance)};self.onunhandledrejection=e=>{throw e.reason||e};function handleMessage(e){try{if(e.data.cmd==="load"){let messageQueue=[];self.onmessage=e=>messageQueue.push(e);self.startWorker=instance=>{Module=instance;postMessage({"cmd":"loaded"});for(let msg of messageQueue){handleMessage(msg)}self.onmessage=handleMessage};Module["wasmModule"]=e.data.wasmModule;for(const handler of e.data.handlers){Module[handler]=(...args)=>{postMessage({cmd:"callHandler",handler:handler,args:args})}}Module["wasmMemory"]=e.data.wasmMemory;Module["buffer"]=Module["wasmMemory"].buffer;Module["ENVIRONMENT_IS_PTHREAD"]=true;if(typeof e.data.urlOrBlob=="string"){importScripts(e.data.urlOrBlob)}else{var objectUrl=URL.createObjectURL(e.data.urlOrBlob);importScripts(objectUrl);URL.revokeObjectURL(objectUrl)}whisper_factory(Module)}else if(e.data.cmd==="run"){Module["__emscripten_thread_init"](e.data.pthread_ptr,0,0,1);Module["__emscripten_thread_mailbox_await"](e.data.pthread_ptr);Module["establishStackSpace"]();Module["PThread"].receiveObjectTransfer(e.data);Module["PThread"].threadInitTLS();if(!initializedJS){Module["__embind_initialize_bindings"]();initializedJS=true}try{Module["invokeEntryPoint"](e.data.start_routine,e.data.arg)}catch(ex){if(ex!="unwind"){throw ex}}}else if(e.data.cmd==="cancel"){if(Module["_pthread_self"]()){Module["__emscripten_thread_exit"](-1)}}else if(e.data.target==="setimmediate"){}else if(e.data.cmd==="checkMailbox"){if(initializedJS){Module["checkMailbox"]()}}else if(e.data.cmd){err(`worker.js received unknown command ${e.data.cmd}`);err(e.data)}}catch(ex){if(Module["__emscripten_thread_crashed"]){Module["__emscripten_thread_crashed"]()}throw ex}}self.onmessage=handleMessage;

View File

@@ -0,0 +1,26 @@
{
"name": "whisper.cpp",
"version": "@PROJECT_VERSION@",
"description": "Whisper speech recognition",
"main": "whisper.js",
"scripts": {
"test": "echo \"todo: add tests\" && exit 0"
},
"repository": {
"type": "git",
"url": "git+https://github.com/ggerganov/whisper.cpp"
},
"keywords": [
"openai",
"whisper",
"speech-to-text",
"speech-recognition",
"transformer"
],
"author": "Georgi Gerganov",
"license": "MIT",
"bugs": {
"url": "https://github.com/ggerganov/whisper.cpp/issues"
},
"homepage": "https://github.com/ggerganov/whisper.cpp#readme"
}

View File

@@ -0,0 +1,26 @@
{
"name": "whisper.cpp",
"version": "1.5.2",
"description": "Whisper speech recognition",
"main": "whisper.js",
"scripts": {
"test": "echo \"todo: add tests\" && exit 0"
},
"repository": {
"type": "git",
"url": "git+https://github.com/ggerganov/whisper.cpp"
},
"keywords": [
"openai",
"whisper",
"speech-to-text",
"speech-recognition",
"transformer"
],
"author": "Georgi Gerganov",
"license": "MIT",
"bugs": {
"url": "https://github.com/ggerganov/whisper.cpp/issues"
},
"homepage": "https://github.com/ggerganov/whisper.cpp#readme"
}

File diff suppressed because one or more lines are too long