To enable word-level timestamp transcription, ensure the following steps are followed: 1. Enable TTS-Aligned Transcription: Set

use_tts_aligned_transcript=True

in your configuration. This synchronizes transcription output with speech timing for supported TTS providers. 2. Use a Supported TTS Provider: Currently, only Cartesia and ElevenLabs support word-level transcription timing. For other providers, alignment is applied at the sentence level. 3. Access Timing Information: Implement a

transcription_node

method in your agent to access timing information. The iterator yields

TimedString

objects, which include

start_time

and

end_time

for each word. Here is an example implementation for accessing timing information:

async def transcription_node(
    self, text: AsyncIterable[str | TimedString], model_settings: ModelSettings
) -> AsyncGenerator[str | TimedString, None]:
    async for chunk in text:
        if isinstance(chunk, TimedString):
            <http://logger.info|logger.info>(f"TimedString: '{chunk}' ({chunk.start_time} - {chunk.end_time})")
        yield chunk

Ensure that your TTS provider explicitly supports this feature. For unsupported providers, alignment will only be at the sentence level. Let me know if you need further assistance! Sources: Text and transcriptions | LiveKit Docs | ElevenLabs TTS integration guide | LiveKit Docs | Cartesia TTS integration guide | LiveKit Docs | livekit-agents/livekit/agents/tts/tts.py