From 89ed593d2025a10b085a948b045c11f49e4089a9 Mon Sep 17 00:00:00 2001
From: Fabien Benetou <fabien-services@benetou.fr>
Date: Sun, 16 Jun 2024 22:51:42 +0200
Subject: [PATCH] working on Quest now too via Vosk

---
 index.html              | 96 ++++++++++++++++++++++++++++++++++-------
 recognizer-processor.js | 39 +++++++++++++++++
 2 files changed, 120 insertions(+), 15 deletions(-)
 create mode 100644 recognizer-processor.js
diff --git a/index.html b/index.html
index 0dbe654..0646bb1 100644
--- a/index.html
+++ b/index.html
@@ -9,11 +9,12 @@
     <script src='dependencies/webdav.js'></script> 
     <script src='jxr-core.js?1234'></script> 
     <script src='jxr-postitnote.js?13235'></script> 
+    <script type="application/javascript" src="https://cdn.jsdelivr.net/npm/vosk-browser@0.0.8/dist/vosk.js"></script>
   </head>
   <body>
 
 <script>
-
+ 
 //________________________________________________________________
 const endpointDomainOrIP = '192.168.0.129' // note that if the certificate is NOT proper, then consider opening it first to accept it on device
 	// e.g https://hmd.link/?https://192.168.0.129:8443/
@@ -55,23 +56,88 @@ function getAllPrimitives(){
 		.map( (i,j) => i )
 } // adapted from https://git.benetou.fr/utopiah/text-code-xr-engine/src/commit/0e1f297ec0cd17b0356811dfa0ab55f1e2629e7c/index.html#L2101
 
+// should test first
 const SpeechRecognition = window.webkitSpeechRecognition; 
-recognizer = new SpeechRecognition();
-recognizer.interimResults = true;
-recognizer.continuous = true;
-// does not work recognizer.lang = 'fr-FR';
-recognizer.lang = 'en-US';
+//(SpeechRecognition) ? console.log('should switch back to native WebSpeech API from speech branch') : console.log('polyfilling WebSpeech API')
+(SpeechRecognition) ? nativeSpeechRecognition( parseSpeech ) : startVoiceRecognition( parseSpeech )
+
+function nativeSpeechRecognition(callbackOnComplete){
+	recognizer = new SpeechRecognition();
+	recognizer.interimResults = true;
+	recognizer.continuous = true;
+	// does not work recognizer.lang = 'fr-FR';
+	recognizer.lang = 'en-US';
+
+	recognizer.onresult = (event) => {
+	   let result = event.results[event.resultIndex]
+	   if (result.isFinal) {
+		console.log('You said: ' + result[0].transcript ) 
+
+		let speechContent = result[0].transcript
+		callbackOnComplete( speechContent )
+	   }
+	}
+}
+
+
+// recognizer.start();
+
+async function startVoiceRecognition( callbackOnComplete ) {
+/* requires
+    recognizer-processor.js
+    https://fabien.benetou.fr/pub/home/future_of_text_demo/engine/vosk-browser/vosk-model-small-en-us-0.15.tar.gz
+
+    from https://github.com/ccoreilly/vosk-browser/tree/master/examples/modern-vanilla
+*/
+
+
+    const channel = new MessageChannel();
+    // const model = await Vosk.createModel('model.tar.gz');
+    const model = await Vosk.createModel('https://fabien.benetou.fr/pub/home/future_of_text_demo/engine/vosk-browser/vosk-model-small-en-us-0.15.tar.gz');
+    model.registerPort(channel.port1);
+
+    const sampleRate = 48000;
+    
+    const recognizer = new model.KaldiRecognizer(sampleRate);
+    recognizer.setWords(true);
+
+    recognizer.on("result", (message) => {
+        const result = message.result;
+        if (result) console.log(JSON.stringify(result, null, 2));
+        
+	callbackOnComplete( result.text )
+    });
+    recognizer.on("partialresult", (message) => {
+        const partial = message.result.partial;
+
+	if (partial) console.log(partial)
+    });
+    
+    const mediaStream = await navigator.mediaDevices.getUserMedia({
+        video: false,
+        audio: {
+            echoCancellation: true,
+            noiseSuppression: true,
+            channelCount: 1,
+            sampleRate
+        },
+    });
+    
+    const audioContext = new AudioContext();
+    await audioContext.audioWorklet.addModule('recognizer-processor.js')
+
+    const recognizerProcessor = new AudioWorkletNode(audioContext, 'recognizer-processor', { channelCount: 1, numberOfInputs: 1, numberOfOutputs: 1 });
+    recognizerProcessor.port.postMessage({action: 'init', recognizerId: recognizer.id}, [ channel.port2 ])
+    recognizerProcessor.connect(audioContext.destination);
+    
+    const source = audioContext.createMediaStreamSource(mediaStream);
+    source.connect(recognizerProcessor);
+}
 
 const aframeprimitives = getAllPrimitives()
 const speechactions = [ "add", "apply", "delete", "clone", "model", "undo" ]
 const speechcustomcomponents = [ "target", "teleporter" ]
-
-recognizer.onresult = (event) => {
-   let result = event.results[event.resultIndex]
-   if (result.isFinal) {
-	console.log('You said: ' + result[0].transcript ) 
-
-	let speechContent = result[0].transcript
+function parseSpeech( speechContent ) {
 	let latest = addedContent.at(-1) 
 	let cmd_words = speechContent.split(" ").map( i => i.toLowerCase() )
 	let el
@@ -145,9 +211,9 @@ recognizer.onresult = (event) => {
 		addedContent.push( addNewNoteAsPostItNote(speechContent, "0 1.2 -.5") )
 			// could become jxr code proper later, also allowing to re-execute a command again
 	} 
-   }
 }
-recognizer.start();
+
+//------------------------------------------------------------------------------------------------------------------------------
 
 var forceXaxis
 // setInterval( _ => console.log(forceXaxis), 1000)
diff --git a/recognizer-processor.js b/recognizer-processor.js
new file mode 100644
index 0000000..41be1f6
--- /dev/null
+++ b/recognizer-processor.js
@@ -0,0 +1,39 @@
+class RecognizerAudioProcessor extends AudioWorkletProcessor {
+    constructor(options) {
+        super(options);
+        
+        this.port.onmessage = this._processMessage.bind(this);
+    }
+    
+    _processMessage(event) {
+        // console.debug(`Received event ${JSON.stringify(event.data, null, 2)}`);
+        if (event.data.action === "init") {
+            this._recognizerId = event.data.recognizerId;
+            this._recognizerPort = event.ports[0];
+        }
+    }
+    
+    process(inputs, outputs, parameters) {
+        const data = inputs[0][0];
+        if (this._recognizerPort && data) {
+            // AudioBuffer samples are represented as floating point numbers between -1.0 and 1.0 whilst
+            // Kaldi expects them to be between -32768 and 32767 (the range of a signed int16)
+            const audioArray = data.map((value) => value * 0x8000);
+        
+            this._recognizerPort.postMessage(
+                {
+                    action: "audioChunk",
+                    data: audioArray,
+                    recognizerId: this._recognizerId,
+                    sampleRate, // Part of AudioWorkletGlobalScope
+                },
+                {
+                    transfer: [audioArray.buffer],
+                }
+            );
+        }
+        return true;
+    }
+}
+
+registerProcessor('recognizer-processor', RecognizerAudioProcessor)
\ No newline at end of file
