fix: Speech gate - only send recording if actual speech detected
- VAD_SPEECH_THRESHOLD_DB = -35 (louder than silence threshold) - Needs 300ms of speech before counting as real speech - Recording discarded if only background noise detected - Prevents sending garbage to Whisper in conversation mode Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
parent
4bbc6f7787
commit
acc13aef6b
|
|
@ -42,6 +42,8 @@ const AUDIO_ENCODING = 'audio/wav';
|
|||
// VAD (Voice Activity Detection) — Stille-Erkennung
|
||||
const VAD_SILENCE_THRESHOLD_DB = -45; // dB unter dem als "Stille" gilt
|
||||
const VAD_SILENCE_DURATION_MS = 1800; // ms Stille bevor Auto-Stop
|
||||
const VAD_SPEECH_THRESHOLD_DB = -35; // dB ueber dem als "Sprache" gilt (Sprach-Gate)
|
||||
const VAD_SPEECH_MIN_MS = 300; // ms Sprache bevor Aufnahme zaehlt
|
||||
|
||||
// --- Audio-Service ---
|
||||
|
||||
|
|
@ -61,6 +63,10 @@ class AudioService {
|
|||
private preloadedSound: Sound | null = null;
|
||||
private preloadedPath: string = '';
|
||||
|
||||
// Sprach-Gate: Aufnahme erst senden wenn tatsaechlich gesprochen wurde
|
||||
private speechDetected: boolean = false;
|
||||
private speechStartTime: number = 0;
|
||||
|
||||
// VAD State
|
||||
private vadEnabled: boolean = false;
|
||||
private lastSpeechTime: number = 0;
|
||||
|
|
@ -128,7 +134,21 @@ class AudioService {
|
|||
const db = e.currentMetering ?? -160;
|
||||
this.meterListeners.forEach(cb => cb(db));
|
||||
|
||||
// VAD: Stille erkennen
|
||||
// Sprach-Gate: Erkennen ob tatsaechlich gesprochen wird
|
||||
if (db > VAD_SPEECH_THRESHOLD_DB) {
|
||||
if (!this.speechDetected && this.speechStartTime === 0) {
|
||||
this.speechStartTime = Date.now();
|
||||
}
|
||||
if (this.speechStartTime > 0 && Date.now() - this.speechStartTime >= VAD_SPEECH_MIN_MS) {
|
||||
this.speechDetected = true;
|
||||
}
|
||||
} else {
|
||||
if (!this.speechDetected) {
|
||||
this.speechStartTime = 0; // Reset wenn noch nicht als Sprache erkannt
|
||||
}
|
||||
}
|
||||
|
||||
// VAD: Stille erkennen (nur wenn Sprache erkannt wurde)
|
||||
if (this.vadEnabled) {
|
||||
if (db > VAD_SILENCE_THRESHOLD_DB) {
|
||||
this.lastSpeechTime = Date.now();
|
||||
|
|
@ -138,6 +158,8 @@ class AudioService {
|
|||
|
||||
this.recordingStartTime = Date.now();
|
||||
this.lastSpeechTime = Date.now();
|
||||
this.speechDetected = false;
|
||||
this.speechStartTime = 0;
|
||||
this.setState('recording');
|
||||
|
||||
// VAD aktivieren
|
||||
|
|
@ -180,6 +202,15 @@ class AudioService {
|
|||
this.recorder.removeRecordBackListener();
|
||||
|
||||
const durationMs = Date.now() - this.recordingStartTime;
|
||||
const hadSpeech = this.speechDetected;
|
||||
|
||||
// Sprach-Gate: Wenn keine Sprache erkannt → Aufnahme verwerfen
|
||||
if (!hadSpeech) {
|
||||
RNFS.unlink(this.recordingPath).catch(() => {});
|
||||
this.setState('idle');
|
||||
console.log('[Audio] Aufnahme verworfen — keine Sprache erkannt (nur Umgebungsgeraeusche)');
|
||||
return null;
|
||||
}
|
||||
|
||||
// Audio-Datei als Base64 lesen
|
||||
const base64Data = await RNFS.readFile(this.recordingPath, 'base64');
|
||||
|
|
@ -188,7 +219,7 @@ class AudioService {
|
|||
RNFS.unlink(this.recordingPath).catch(() => {});
|
||||
|
||||
this.setState('idle');
|
||||
console.log(`[Audio] Aufnahme beendet (${durationMs}ms, ${Math.round(base64Data.length / 1024)}KB)`);
|
||||
console.log(`[Audio] Aufnahme beendet (${durationMs}ms, ${Math.round(base64Data.length / 1024)}KB, Sprache erkannt)`);
|
||||
|
||||
return {
|
||||
base64: base64Data,
|
||||
|
|
|
|||
Loading…
Reference in New Issue