`use_tts_aligned_transcript` is enabled but no agent transcript was returned from tts

[DeoLeung]

I enable use_tts_aligned_transcript

and my tts actually do

if words := info['words']:
  timed_words: list[TimedString] = []
  for word in words:
    timed_words.append(
      TimedString(
        word['word'],
        start_time=word['startTime'],
        end_time=word['endTime'],
      )
    )
  output_emitter.push_timed_transcript(timed_words)

i overwrite the transcription_node and do see the TimedString chunk, but i still got the warning, the only thing i can think of is for the first sentence audio comes first before the timed words

sentence start -> audio chunks -> sentence end(timed words) -> sentence start -> ...

my question is, in this case, does use_tts_aligned_transcript helps? will it pick up the timed words once they are available, or just got ignored?

also i octionally got my tts stop at the middle ( from server log i can see it recieved all the audio chunks), so i wonder whether this setting may be the cause

---

i think the warning may comes from the say i use in tool calls with cached audio bytes

audio_bytes = await ctx.userdata.cache.get(key)
    if audio_bytes is None:
      audio_bytes = await session.tts.synthesize(text, return_bytes=True)
      await ctx.userdata.cache.set(
        key,
        audio_bytes,
      )
    session.say(
      text,
      audio=say_audio(audio_bytes),
      allow_interruptions=allow_interruptions,
      add_to_chat_ctx=False,
    )

so any suggestion about improving the aligment in session.say?

[longcw]

this warning message should only be logged from agent generation, so may not related to the session.say.

did you yield the timed string back from the transcription_node, or any unexpected behaviors besides the warning?

[DeoLeung]

this warning message should only be logged from agent generation, so may not related to the session.say.

did you yield the timed string back from the transcription_node, or any unexpected behaviors besides the warning?

so far just the warning, i will debug it later to see what exactly happened,

btw , do i need to implement the transcription_node? i see it should already have it by default(i already yield Timedstring fom tts)? https://docs.livekit.io/agents/build/text/#tts-aligned-transcriptions

[longcw]

do i need to implement the transcription_node?

only if you want to read the timestamps or modify the transcripts