PowerOn/gateway
2
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

feat: add speaker-hint debug flag and improve TTS diagnostics

Browse source 
Made-with: Cursor
This commit is contained in:
patrick-motsch 2026-02-26 09:26:59 +01:00
parent 0cd8e9ebfa
commit 02002f3576
1 changed files with 44 additions and 19 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

63
modules/features/teamsbot/service.py
View file

@ -444,13 +444,14 @@ class TeamsbotService:
logger.info(f"[AudioChunk] STT result: {text[:80]}...") logger.info(f"[AudioChunk] STT result: {text[:80]}...")
await self._processTranscript( await self._processTranscript(
sessionId=sessionId, sessionId=sessionId,
speaker=resolvedSpeaker, speaker=resolvedSpeaker["speaker"],
text=text, text=text,
isFinal=True, isFinal=True,
interface=interface, interface=interface,
voiceInterface=voiceInterface, voiceInterface=voiceInterface,
websocket=websocket, websocket=websocket,
source="audioCapture", source="audioCapture",
speakerResolvedFromHint=resolvedSpeaker["speakerResolvedFromHint"],
) )
except Exception as e: except Exception as e:
logger.error(f"[AudioChunk] STT error for session {sessionId}: {type(e).__name__}: {e}") logger.error(f"[AudioChunk] STT error for session {sessionId}: {type(e).__name__}: {e}")
@ -473,19 +474,22 @@ class TeamsbotService:
if len(self._recentSpeakerHints) > 20: if len(self._recentSpeakerHints) > 20:
self._recentSpeakerHints = self._recentSpeakerHints[-20:] self._recentSpeakerHints = self._recentSpeakerHints[-20:]
def _resolveSpeakerForAudioCapture(self) -> str: def _resolveSpeakerForAudioCapture(self) -> Dict[str, Any]:
"""Best-effort speaker name for audio chunks using recent caption hints.""" """Best-effort speaker name for audio chunks using recent caption hints."""
if not self._recentSpeakerHints: if not self._recentSpeakerHints:
return "Meeting Audio" return {"speaker": "Meeting Audio", "speakerResolvedFromHint": False}
nowTs = time.time() nowTs = time.time()
# Prefer very recent hints to reduce wrong attribution # Prefer very recent hints to reduce wrong attribution
for hint in reversed(self._recentSpeakerHints): for hint in reversed(self._recentSpeakerHints):
hintAge = nowTs - hint.get("timestamp", 0) hintAge = nowTs - hint.get("timestamp", 0)
if hintAge <= 15: if hintAge <= 15:
return hint.get("speaker", "Meeting Audio") return {
"speaker": hint.get("speaker", "Meeting Audio"),
"speakerResolvedFromHint": True,
}
return "Meeting Audio" return {"speaker": "Meeting Audio", "speakerResolvedFromHint": False}
async def _processTranscript( async def _processTranscript(
self, self,
@ -497,6 +501,7 @@ class TeamsbotService:
voiceInterface, voiceInterface,
websocket: WebSocket, websocket: WebSocket,
source: str = "caption", source: str = "caption",
speakerResolvedFromHint: Optional[bool] = None,
): ):
"""Process a transcript segment from captions or chat messages. """Process a transcript segment from captions or chat messages.
@ -601,6 +606,12 @@ class TeamsbotService:
"confidence": 1.0, "confidence": 1.0,
"timestamp": getIsoTimestamp(), "timestamp": getIsoTimestamp(),
"isContinuation": isContinuation, "isContinuation": isContinuation,
"source": source,
"speakerResolvedFromHint": (
speakerResolvedFromHint
if speakerResolvedFromHint is not None
else False
),
}) })
# Check if AI analysis should be triggered (only for final transcripts) # Check if AI analysis should be triggered (only for final transcripts)
@ -837,25 +848,39 @@ class TeamsbotService:
# 4a: Voice response (TTS -> Audio to bot) # 4a: Voice response (TTS -> Audio to bot)
if sendVoice: if sendVoice:
try: try:
logger.info(
f"Session {sessionId}: TTS requested (websocket_available={websocket is not None})"
)
ttsResult = await voiceInterface.textToSpeech( ttsResult = await voiceInterface.textToSpeech(
text=speechResult.responseText, text=speechResult.responseText,
languageCode=self.config.language, languageCode=self.config.language,
voiceName=self.config.voiceId voiceName=self.config.voiceId
) )
if ttsResult and isinstance(ttsResult, dict): if not ttsResult or not isinstance(ttsResult, dict):
audioContent = ttsResult.get("audioContent") raise RuntimeError("TTS returned invalid result payload")
if audioContent and websocket:
await websocket.send_text(json.dumps({ if ttsResult.get("success") is False:
"type": "playAudio", raise RuntimeError(f"TTS backend error: {ttsResult.get('error', 'unknown')}")
"sessionId": sessionId,
"audio": { audioContent = ttsResult.get("audioContent")
"data": base64.b64encode(audioContent if isinstance(audioContent, bytes) else audioContent.encode()).decode(), if not audioContent:
"format": "mp3", raise RuntimeError("TTS returned no audioContent")
},
})) if websocket:
elif audioContent and not websocket: await websocket.send_text(json.dumps({
logger.info(f"TTS audio generated for session {sessionId} (HTTP mode - no WebSocket for playback)") "type": "playAudio",
"sessionId": sessionId,
"audio": {
"data": base64.b64encode(audioContent if isinstance(audioContent, bytes) else audioContent.encode()).decode(),
"format": "mp3",
},
}))
logger.info(f"Session {sessionId}: TTS audio dispatched to bot")
else:
logger.warning(
f"Session {sessionId}: TTS audio generated but cannot be played (bot websocket unavailable, likely fallback mode)"
)
except Exception as ttsErr: except Exception as ttsErr:
logger.warning(f"TTS failed for session {sessionId}: {ttsErr}") logger.warning(f"TTS failed for session {sessionId}: {ttsErr}")
if not sendChat: if not sendChat: