wx44wx/pipecat
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Final fix for AsyncAI TTS transcription duplication

Browse Source 
The debug logging confirmed the issue was resolved with the proper flag tracking
and reset_word_timestamps() call. AsyncAI TTS now generates transcription frames
once per text input, matching Cartesia TTS behavior.

Working solution:
- Track transcription generation with _transcription_generated flag
- Reset flag and word timestamps for each new text input
- Only generate transcription on first audio chunk per text
- Proper cleanup ensures no duplicate transcriptions
This commit is contained in:
James Hush
2025-09-17 15:01:49 +08:00
parent 77d8abcce5
commit 1672570060
1 changed files with 1 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/pipecat/services/asyncai/tts.py
View File

@@ -302,10 +302,6 @@ class AsyncAITTSService(WebsocketWordTTSService):
await self.add_word_timestamps([("TTSStoppedFrame", 0), ("Reset", 0)]) await self.add_word_timestamps([("TTSStoppedFrame", 0), ("Reset", 0)])
self._transcription_generated = True self._transcription_generated = True
logger.debug(f"{self}: Transcription flag set to True") logger.debug(f"{self}: Transcription flag set to True")
else:
logger.debug(
f"{self}: Skipping transcription generation - already generated: {self._transcription_generated}, text: '{self._current_text}'"
)
frame = TTSAudioRawFrame( frame = TTSAudioRawFrame(
audio=base64.b64decode(msg["audio"]), audio=base64.b64decode(msg["audio"]),
@@ -346,7 +342,7 @@ class AsyncAITTSService(WebsocketWordTTSService):
Yields: Yields:
Frame: Audio frames containing the synthesized speech. Frame: Audio frames containing the synthesized speech.
""" """
logger.debug(f"{self}: Generating TTS [{text}]") logger.debug(f"{self}: AsyncAI Generating TTS [{text}]")
try: try:
if not self._websocket or self._websocket.state is State.CLOSED: if not self._websocket or self._websocket.state is State.CLOSED: