110 lines
3.0 KiB
JavaScript
110 lines
3.0 KiB
JavaScript
// Copyright (c) 2023-2024 Xiaomi Corporation (authors: Fangjun Kuang)
|
|
//
|
|
const portAudio = require('naudiodon2');
|
|
// console.log(portAudio.getDevices());
|
|
|
|
const sherpa_onnx = require('sherpa-onnx-node');
|
|
|
|
function createRecognizer() {
|
|
// Please download test files from
|
|
// https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models
|
|
const config = {
|
|
'featConfig': {
|
|
'sampleRate': 16000,
|
|
'featureDim': 80,
|
|
},
|
|
'modelConfig': {
|
|
'whisper': {
|
|
'encoder': './sherpa-onnx-whisper-tiny.en/tiny.en-encoder.int8.onnx',
|
|
'decoder': './sherpa-onnx-whisper-tiny.en/tiny.en-decoder.int8.onnx',
|
|
},
|
|
'tokens': './sherpa-onnx-whisper-tiny.en/tiny.en-tokens.txt',
|
|
'numThreads': 2,
|
|
'provider': 'cpu',
|
|
'debug': 1,
|
|
}
|
|
};
|
|
|
|
return new sherpa_onnx.OfflineRecognizer(config);
|
|
}
|
|
|
|
function createVad() {
|
|
// please download silero_vad.onnx from
|
|
// https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/silero_vad.onnx
|
|
const config = {
|
|
sileroVad: {
|
|
model: './silero_vad.onnx',
|
|
threshold: 0.5,
|
|
minSpeechDuration: 0.25,
|
|
minSilenceDuration: 0.5,
|
|
windowSize: 512,
|
|
},
|
|
sampleRate: 16000,
|
|
debug: true,
|
|
numThreads: 1,
|
|
};
|
|
|
|
const bufferSizeInSeconds = 60;
|
|
|
|
return new sherpa_onnx.Vad(config, bufferSizeInSeconds);
|
|
}
|
|
|
|
const recognizer = createRecognizer();
|
|
const vad = createVad();
|
|
|
|
const bufferSizeInSeconds = 30;
|
|
const buffer =
|
|
new sherpa_onnx.CircularBuffer(bufferSizeInSeconds * vad.config.sampleRate);
|
|
|
|
const ai = new portAudio.AudioIO({
|
|
inOptions: {
|
|
channelCount: 1,
|
|
closeOnError: true, // Close the stream if an audio error is detected, if
|
|
// set false then just log the error
|
|
deviceId: -1, // Use -1 or omit the deviceId to select the default device
|
|
sampleFormat: portAudio.SampleFormatFloat32,
|
|
sampleRate: vad.config.sampleRate
|
|
}
|
|
});
|
|
|
|
let printed = false;
|
|
let index = 0;
|
|
ai.on('data', data => {
|
|
const windowSize = vad.config.sileroVad.windowSize;
|
|
buffer.push(new Float32Array(data.buffer));
|
|
while (buffer.size() > windowSize) {
|
|
const samples = buffer.get(buffer.head(), windowSize);
|
|
buffer.pop(windowSize);
|
|
vad.acceptWaveform(samples);
|
|
}
|
|
|
|
while (!vad.isEmpty()) {
|
|
const segment = vad.front();
|
|
vad.pop();
|
|
const stream = recognizer.createStream();
|
|
stream.acceptWaveform({
|
|
samples: segment.samples,
|
|
sampleRate: recognizer.config.featConfig.sampleRate
|
|
});
|
|
recognizer.decode(stream);
|
|
const r = recognizer.getResult(stream);
|
|
if (r.text.length > 0) {
|
|
const text = r.text.toLowerCase().trim();
|
|
console.log(`${index}: ${text}`);
|
|
|
|
const filename = `${index}-${text}-${
|
|
new Date()
|
|
.toLocaleTimeString('en-US', {hour12: false})
|
|
.split(' ')[0]}.wav`;
|
|
sherpa_onnx.writeWave(
|
|
filename,
|
|
{samples: segment.samples, sampleRate: vad.config.sampleRate});
|
|
|
|
index += 1;
|
|
}
|
|
}
|
|
});
|
|
|
|
ai.start();
|
|
console.log('Started! Please speak')
|