mirror of
https://github.com/R0m1k3/EveFlow.git
synced 2026-10-11 17:29:03 +02:00
fix: implement direct WAV recording to resolve transcription API media decoding errors
This commit is contained in:
Michael committed
1 parent
cf52a610d4
commit
57690962e1
1 file changed
+157
-52
+157
-52
@@ -64,8 +64,20 @@ export class AudioService {
|
||||
private sttApiUrl = 'http://127.0.0.1:8000/v1';
|
||||
private sttApiKey = '';
|
||||
private sttModel = 'Qwen/Qwen3-ASR-0.6B';
|
||||
private mediaRecorder: MediaRecorder | null = null;
|
||||
private audioChunks: Blob[] = [];
|
||||
|
||||
// Propriétés pour l'enregistrement WAV direct
|
||||
private audioContext: AudioContext | null = null;
|
||||
private scriptProcessor: ScriptProcessorNode | null = null;
|
||||
private micStream: MediaStream | null = null;
|
||||
private leftChannel: Float32Array[] = [];
|
||||
private recordingLength = 0;
|
||||
private sampleRate = 44100;
|
||||
|
||||
// Callbacks pour l'écoute
|
||||
private activeOnResult: ((text: string) => void) | null = null;
|
||||
private activeOnEnd: (() => void) | null = null;
|
||||
private activeOnError: ((err: string) => void) | null = null;
|
||||
private activeLang = 'fr-FR';
|
||||
|
||||
private ttsApiUrl = 'http://127.0.0.1:8000/v1';
|
||||
private ttsApiKey = '';
|
||||
@@ -584,7 +596,7 @@ export class AudioService {
|
||||
}
|
||||
}
|
||||
|
||||
// Démarrer l'écoute vocale API via MediaRecorder
|
||||
// Démarrer l'écoute vocale API via AudioContext (WAV Recorder)
|
||||
private async startListeningAPI(
|
||||
onResult: (text: string) => void,
|
||||
onStart: () => void,
|
||||
@@ -596,59 +608,156 @@ export class AudioService {
|
||||
this.stopListening();
|
||||
}
|
||||
|
||||
// Assigner les callbacks pour y avoir accès à l'arrêt
|
||||
this.activeOnResult = onResult;
|
||||
this.activeOnEnd = onEnd;
|
||||
this.activeOnError = onError;
|
||||
this.activeLang = lang;
|
||||
|
||||
try {
|
||||
const stream = await navigator.mediaDevices.getUserMedia({ audio: true });
|
||||
this.audioChunks = [];
|
||||
this.micStream = stream;
|
||||
this.leftChannel = [];
|
||||
this.recordingLength = 0;
|
||||
|
||||
let options = {};
|
||||
if (MediaRecorder.isTypeSupported('audio/webm;codecs=opus')) {
|
||||
options = { mimeType: 'audio/webm;codecs=opus' };
|
||||
} else if (MediaRecorder.isTypeSupported('audio/webm')) {
|
||||
options = { mimeType: 'audio/webm' };
|
||||
}
|
||||
const AudioCtx = window.AudioContext || (window as any).webkitAudioContext;
|
||||
this.audioContext = new AudioCtx();
|
||||
this.sampleRate = this.audioContext.sampleRate;
|
||||
|
||||
const source = this.audioContext.createMediaStreamSource(stream);
|
||||
// bufferSize de 2048, 1 canal d'entrée, 1 de sortie
|
||||
this.scriptProcessor = this.audioContext.createScriptProcessor(2048, 1, 1);
|
||||
|
||||
this.scriptProcessor.onaudioprocess = (e) => {
|
||||
if (!this.isListeningActive) return;
|
||||
const inputData = e.inputBuffer.getChannelData(0);
|
||||
// Cloner les échantillons
|
||||
this.leftChannel.push(new Float32Array(inputData));
|
||||
this.recordingLength += inputData.length;
|
||||
};
|
||||
|
||||
source.connect(this.scriptProcessor);
|
||||
this.scriptProcessor.connect(this.audioContext.destination);
|
||||
|
||||
this.mediaRecorder = new MediaRecorder(stream, options);
|
||||
this.isListeningActive = true;
|
||||
|
||||
this.mediaRecorder.ondataavailable = (event) => {
|
||||
if (event.data && event.data.size > 0) {
|
||||
this.audioChunks.push(event.data);
|
||||
}
|
||||
};
|
||||
|
||||
this.mediaRecorder.onstart = () => {
|
||||
onStart();
|
||||
};
|
||||
|
||||
this.mediaRecorder.onstop = async () => {
|
||||
this.isListeningActive = false;
|
||||
onEnd();
|
||||
|
||||
// Stopper toutes les pistes pour éteindre le voyant du microphone
|
||||
stream.getTracks().forEach(track => track.stop());
|
||||
|
||||
if (this.audioChunks.length === 0) {
|
||||
onError("Aucune donnée audio capturée");
|
||||
return;
|
||||
}
|
||||
|
||||
const audioBlob = new Blob(this.audioChunks, { type: this.mediaRecorder?.mimeType || 'audio/webm' });
|
||||
|
||||
try {
|
||||
const text = await this.transcribeAudioWithAPI(audioBlob, lang);
|
||||
onResult(text);
|
||||
} catch (err: any) {
|
||||
onError(err.message || "Erreur de transcription de l'API");
|
||||
}
|
||||
};
|
||||
|
||||
this.mediaRecorder.start();
|
||||
onStart();
|
||||
} catch (err: any) {
|
||||
this.isListeningActive = false;
|
||||
this.logError("STT", "Impossible d'accéder au microphone", err);
|
||||
onError(err.message || "Impossible d'accéder au microphone.");
|
||||
}
|
||||
}
|
||||
|
||||
// Arrêter proprement la capture AudioContext et exporter en WAV
|
||||
private stopListeningAPI() {
|
||||
if (!this.isListeningActive) return;
|
||||
this.isListeningActive = false;
|
||||
|
||||
// Déconnecter le processeur
|
||||
if (this.scriptProcessor) {
|
||||
this.scriptProcessor.disconnect();
|
||||
this.scriptProcessor.onaudioprocess = null;
|
||||
this.scriptProcessor = null;
|
||||
}
|
||||
|
||||
// Arrêter le flux micro
|
||||
if (this.micStream) {
|
||||
this.micStream.getTracks().forEach(track => track.stop());
|
||||
this.micStream = null;
|
||||
}
|
||||
|
||||
// Fermer l'audio context
|
||||
if (this.audioContext) {
|
||||
this.audioContext.close();
|
||||
this.audioContext = null;
|
||||
}
|
||||
|
||||
const onEndCallback = this.activeOnEnd;
|
||||
const onResultCallback = this.activeOnResult;
|
||||
const onErrorCallback = this.activeOnError;
|
||||
const lang = this.activeLang;
|
||||
|
||||
if (onEndCallback) onEndCallback();
|
||||
|
||||
if (this.recordingLength === 0) {
|
||||
if (onErrorCallback) onErrorCallback("Aucune donnée audio capturée");
|
||||
return;
|
||||
}
|
||||
|
||||
// Exporter le tampon audio en fichier WAV
|
||||
const wavBlob = this.exportWAV(this.leftChannel, this.recordingLength, this.sampleRate);
|
||||
|
||||
this.transcribeAudioWithAPI(wavBlob, lang)
|
||||
.then(text => {
|
||||
if (onResultCallback) onResultCallback(text);
|
||||
})
|
||||
.catch(err => {
|
||||
this.logError("STT", "Erreur de transcription", err);
|
||||
if (onErrorCallback) onErrorCallback(err.message || "Erreur de transcription de l'API");
|
||||
});
|
||||
}
|
||||
|
||||
// Fonctions utilitaires pour générer le conteneur WAV PCM 16-bit
|
||||
private exportWAV(channelData: Float32Array[], recordingLength: number, sampleRate: number): Blob {
|
||||
const buffer = this.mergeBuffers(channelData, recordingLength);
|
||||
const wavBuffer = new ArrayBuffer(44 + buffer.length * 2);
|
||||
const view = new DataView(wavBuffer);
|
||||
|
||||
/* Identifiant RIFF */
|
||||
this.writeString(view, 0, 'RIFF');
|
||||
/* Taille du fichier */
|
||||
view.setUint32(4, 36 + buffer.length * 2, true);
|
||||
/* Type RIFF */
|
||||
this.writeString(view, 8, 'WAVE');
|
||||
/* Identifiant format fmt */
|
||||
this.writeString(view, 12, 'fmt ');
|
||||
/* Longueur du chunk format */
|
||||
view.setUint32(16, 16, true);
|
||||
/* Format d'encodage (1 = PCM non compressé) */
|
||||
view.setUint16(20, 1, true);
|
||||
/* Nombre de canaux (1 = Mono) */
|
||||
view.setUint16(22, 1, true);
|
||||
/* Fréquence d'échantillonnage */
|
||||
view.setUint32(24, sampleRate, true);
|
||||
/* Débit d'octets (sampleRate * blockAlign) */
|
||||
view.setUint32(28, sampleRate * 2, true);
|
||||
/* Block align (1 canal * 2 octets par échantillon) */
|
||||
view.setUint16(32, 2, true);
|
||||
/* Bits par échantillon (16 bits) */
|
||||
view.setUint16(34, 16, true);
|
||||
/* Identifiant de données data */
|
||||
this.writeString(view, 36, 'data');
|
||||
/* Longueur des données audio */
|
||||
view.setUint32(40, buffer.length * 2, true);
|
||||
|
||||
this.floatTo16BitPCM(view, 44, buffer);
|
||||
|
||||
return new Blob([view], { type: 'audio/wav' });
|
||||
}
|
||||
|
||||
private mergeBuffers(channelData: Float32Array[], recordingLength: number): Float32Array {
|
||||
const result = new Float32Array(recordingLength);
|
||||
let offset = 0;
|
||||
for (let i = 0; i < channelData.length; i++) {
|
||||
result.set(channelData[i], offset);
|
||||
offset += channelData[i].length;
|
||||
}
|
||||
return result;
|
||||
}
|
||||
|
||||
private writeString(view: DataView, offset: number, string: string) {
|
||||
for (let i = 0; i < string.length; i++) {
|
||||
view.setUint8(offset + i, string.charCodeAt(i));
|
||||
}
|
||||
}
|
||||
|
||||
private floatTo16BitPCM(output: DataView, offset: number, input: Float32Array) {
|
||||
for (let i = 0; i < input.length; i++, offset += 2) {
|
||||
const s = Math.max(-1, Math.min(1, input[i]));
|
||||
output.setInt16(offset, s < 0 ? s * 0x8000 : s * 0x7FFF, true);
|
||||
}
|
||||
}
|
||||
|
||||
// Requête API Speech-To-Text compatible OpenAI /v1/audio/transcriptions
|
||||
private async transcribeAudioWithAPI(audioBlob: Blob, lang: string): Promise<string> {
|
||||
if (!this.sttApiUrl) {
|
||||
@@ -666,9 +775,8 @@ export class AudioService {
|
||||
}
|
||||
|
||||
const formData = new FormData();
|
||||
const ext = audioBlob.type.includes('wav') ? 'wav' : 'webm';
|
||||
formData.append('file', audioBlob, `audio.${ext}`);
|
||||
formData.append('model', this.sttModel || 'qwen3-asr');
|
||||
formData.append('file', audioBlob, 'audio.wav');
|
||||
formData.append('model', this.sttModel || 'Qwen/Qwen3-ASR-0.6B');
|
||||
|
||||
if (lang) {
|
||||
const isoLang = lang.split('-')[0];
|
||||
@@ -756,10 +864,7 @@ export class AudioService {
|
||||
// Arrêter l'écoute vocale
|
||||
public stopListening() {
|
||||
if (this.sttProvider === 'qwen3-asr') {
|
||||
if (this.mediaRecorder && this.mediaRecorder.state !== 'inactive') {
|
||||
this.mediaRecorder.stop();
|
||||
this.isListeningActive = false;
|
||||
}
|
||||
this.stopListeningAPI();
|
||||
} else {
|
||||
if (this.recognition && this.isListeningActive) {
|
||||
this.recognition.stop();
|
||||
|
||||
Reference in new issue
Block a user