Add documents and optimize build

src/kaldi.patch

@@ -39,3 +39,15 @@ index c3a122281..71d37256d 100644
      TaskSequencer<IvectorExtractorComputeDerivedVarsClass> sequencer(
          sequencer_opts);
      for (int32 i = 0; i < NumGauss(); i++)
+diff --git a/src/base/kaldi-types.h b/src/base/kaldi-types.h
+index 7ebf4f8..2f5979e 100644
+--- a/src/base/kaldi-types.h
++++ b/src/base/kaldi-types.h
+@@ -20,6 +20,7 @@
+ 
+ #ifndef KALDI_BASE_KALDI_TYPES_H_
+ #define KALDI_BASE_KALDI_TYPES_H_ 1
++#define KALDI_DOUBLEPRECISION 1
+ 
+ namespace kaldi {
+ // TYPEDEFS ..................................................................

src/post.js

@@ -1 +0,0 @@
-window.loadBR = loadBR

src/pre.js

@@ -1,14 +1,14 @@
-var objs =  []
+let objs =  []
 class recognizer extends EventTarget {
   constructor(rec) {
     super()
     this.obj = rec
     objs.push(this)
   }
-  processAudio(buffer) {
-    if(buffer.numberOfChannels < 1) throw Error("Buffer has ",buffer.numberOfChannels, " channel")
-    let data = buffer.getChannelData(0);
-    if(!(data instanceof Float32Array)) throw Error("Channel data isn't a Float32Array");
+  processAudio(ctx) {
+    let ptr = Module._malloc(512);
+
+    this.obj.acceptWaveForm(ptr)
   }
   delete() {
     this.obj.delete()
@@ -32,7 +32,10 @@ class recognizer extends EventTarget {
     this.obj.setMaxAlternatives(alts)
   }
 }
-Module.deleteAll = () => objs.forEach(obj => obj.delete())
+Module.deleteAll = () => {
+  objs.forEach(obj => obj.delete())
+  ctx.close()
+}
 Module.makeModel = async (url, path, id) => {
   let mdl
   try {

src/processor.js

@@ -0,0 +1,23 @@
+registerProcessor("wasmMemCpy", class extends AudioWorkletProcessor {
+  constructor(options) {
+    super(options)
+    this.retval = true
+    this.port.onmessage = (ev) => {
+      switch(ev.cmd) {
+        case "init":
+          this.ptr = ev.ptr
+          this.wasmMem = ev.wasmMem
+          this.recognizerPort = ev.ports[0]
+          break
+        case "deinit":
+          this.retval = false
+          break
+      }
+    }
+  }
+  process(input, output, param) {
+    const data = input[0][0]
+    this.recognizerPort.postMessage("done")
+    return this.retval
+  }
+})

src/recognizer.cc

@@ -1,9 +1,25 @@
 #include "recognizer.h" 
+audioData::audioData(int addr, int len) : addr(reinterpret_cast<float*>(addr)), len(len) {}
 recognizer::recognizer(model* mdl, float sampleRate, int index) : index(index) {
   rec = vosk_recognizer_new(mdl->mdl,sampleRate);
   if(rec == nullptr) {
     throwJS("Unable to initialize recognizer");
-  } 
+    return;
+  }
+  std::thread t{[this](){
+    while(!queue.empty()) {
+      audioData data {queue.front()};
+      queue.pop();
+      switch(vosk_recognizer_accept_waveform_f(rec, data.addr, data.len)) {
+      case 0:
+      fireEv("result", vosk_recognizer_result(rec));
+      break;
+      case 1:
+      fireEv("partialResult", vosk_recognizer_partial_result(rec));
+      }
+    }
+  }};
+  t.detach();
 }
 recognizer::~recognizer() {
   vosk_recognizer_free(rec);
@@ -13,17 +29,8 @@ void recognizer::fireEv(const char *type, const char *content) {
     recognizers[$0].dispatchEvent(new CustomEvent(UTF8ToString($1), {"details" : UTF8ToString($2)}));
   },this->index, type, content);
 }
-void recognizer::acceptWaveForm(float* data, int len) {
-  switch(vosk_recognizer_accept_waveform_f(rec, data, len)) {
-    case 0:
-    fireEv("result", vosk_recognizer_result(rec));
-    break;
-    case 1:
-    fireEv("partialResult", vosk_recognizer_partial_result(rec));
-    break;
-    default:
-    throwJS("acceptWaveForm error (from C++)", true);
-  }
+void recognizer::acceptWaveForm(int addr, int len) {
+  queue.emplace(addr, len);
 }
 void recognizer::setGrm(const std::string& grm) {
   vosk_recognizer_set_grm(rec, grm.c_str());

src/recognizer.h

@@ -5,6 +5,7 @@
 #include <filesystem>
 #include <atomic>
 #include <thread>
+#include <queue>
 
 #include <emscripten/wasmfs.h>
 #include <emscripten/webaudio.h>
@@ -15,12 +16,17 @@
 extern void throwJS(const char* msg, bool err = false);
 namespace fs = std::filesystem;
 
+struct audioData {
+  float* addr{};
+  int len{};
+  audioData(int addr, int len);
+};
 struct recognizer {
   int index{};
   VoskRecognizer* rec{};
   recognizer(model* model, float sampleRate, int index);
   ~recognizer();
-  void acceptWaveForm(float* data, int len);
+  void acceptWaveForm(int addr, int len);
   void fireEv(const char* type, const char* content);
   void setSpkModel(spkModel* model);
   void setGrm(const std::string& grm);

src/vosk.patch

@@ -0,0 +1,16 @@
+diff --git a/src/recognizer.cc b/src/recognizer.cc
+index 1da6e6b..5e0652a 100644
+--- a/src/recognizer.cc
++++ b/src/recognizer.cc
+@@ -384,8 +384,10 @@ bool Recognizer::AcceptWaveform(const float *fdata, int len)
+ {
+     Vector<BaseFloat> wave;
+     wave.Resize(len, kUndefined);
+-    for (int i = 0; i < len; i++)
++    for (int i = 0; i < len; i++) {
+         wave(i) = fdata[i];
++        wave(i) *= 32768;
++    }
+     return AcceptWaveform(wave);
+ }
+