nodejs client example?

[FerLuisxd]

I found there is a python client example to be used on the backend but what about a nodejs one?
I would like to be able to send an audio buffer to the server and get a response

[fquirin]

Hi @FerLuisxd ,

do you handle recording and (optionally) resampling already? Because I don't know how that works in Node.js, but large parts of the SEPIA web-audio lib should still be compatible. Maybe we can pick some parts and put it together for Node.js.

The most important parts are the sepia-stt-socket-client as server interface and the stt-socket-worker that handles the logic like buffering and streaming to the server. Since the worker is part of the web-audio lib, it might be possible to simplify it for stand-alone usage with Node.js.

[Kam1ni]

@FerLuisxd Here you go. Real time transcription of your microphone in Node.js. But this is made for linux. I had no success with the Mic library in windows. So if that doesn't work try Node naudiodon. https://www.npmjs.com/package/naudiodon

These are my dependencies:
"mic": "^2.1.2",
"wav": "^1.0.2",
"ws": "^8.18.1"

index.js:

const Mic = require("./mic");
const Stt = require("./stt");

async function main(){
	let mic = new Mic();
	let stt = new Stt(mic);
	stt.on("data", (msg)=>{
		if (!msg) return;
		console.log(`Transcription: ${msg}`);
	})
	stt.start();
	process.on("SIGINT", ()=>{
		stt.stop();
		process.exit();
	})
}


main();

mic.js:

const PassThrough = require('stream').PassThrough;
const mic = require("mic");
const wav = require("wav");

class Mic extends PassThrough {
	constructor() {
		super();
	}

	start(){
		if (this._proc) return;
		this.wavWriter = new wav.Writer({
			channels:1,
			sampleRate:16000,
			bitDepth:16,
		});
		this._proc = mic({rate:'16000', channels:'1', debug:true});
		this._proc.getAudioStream().pipe(this.wavWriter);
		this.wavWriter.pipe(this);
		this._proc.start();
	}

	stop(){
		if (!this._proc) return;
		this._proc.stop();
		this._proc = null;
	}
}

module.exports = Mic

stt.js

const WebSocket = require("ws");
const EventEmitter = require("./event-emitter")

class Stt extends EventEmitter{
	constructor(audioIn){
		super();
		this.audioIn = audioIn;
		this._byteLength = 0;
		this._shouldClose = false;
		this._started = false;
	}

	async start(){
		if (this._started) return;
		this._started = true;
		this.ws = new WebSocket(`ws://localhost:20741/socket`);
		this.ws.on("open", ()=>{
			console.log("Socket opened");

			this.ws.on("message", (data)=>{
				this._processMessage(data);
			})

			this.ws.send(JSON.stringify({
				type:"welcome",
				data:{
					sampleRate:16000,
					continuous:true,
					langauge:"en-US",
					task:"",
					model:"vosk-model-small-en-us:assistant",
					optimizeFinalResult:true,
					doDebug:false
				},
				client_id:"any",
				access_token:"test1234",
				msg_id:1
			}));

			this.ws.on("close", (handler)=>{
				this.audioIn.stop();
				this.ws.close();
				this._started = false;
			})
		});
	}

	async stop(){
		this.audioIn.stop();
		this.ws.send(JSON.stringify({
			type:"audioend",
			data:{
				byteLength:this._byteLength,
				bufferOrTimeLimit:false
			},
			msg_id:2
		}))
		setTimeout(()=>{
			this.ws.close();
		},3000)
	}

	_processMessage(data){
		let msg = JSON.parse(data);
		//console.log(msg);
		if (msg.type == "welcome"){
			this._processWelcomeMessage(msg);
		}else if (msg.type == "result"){
			this._processResultMessage(msg)
		}else if (msg.type == "response"){
			this._processResponseMessage(msg)
		}
	}
	
	_processWelcomeMessage(msg){
		console.log("Received welcome");
		console.log("Starting microphone")
		this.audioIn.on("data", (chunk)=>{
			this._byteLength += chunk.length;
			this.ws.send(chunk);
		});
		this.audioIn.start();
	}

	_processResultMessage(msg){
		if (!msg.isFinal){
			return;
		}
		this.emit("data", msg.transcript);
	}

	_processResponseMessage(msg){
	}
}


module.exports = Stt

[fquirin]

nice work @Kam1ni . I haven't had time to test it, but it looks very promising :-)