Hi LiveKit team, We are preparing a production deployment of a voice agent built on LiveKit Agents (Python) and are blocked on getting reliable timestamps aligned to TTS output. Our requirement We need: • Start/end timestamps for that transcript, aligned to the actual audio (sufficient at sentence level; word level is a bonus). • This timing needs to come from the TTS engine / LiveKit alignment itself, not from approximate server event timestamps. We are currently using Cartesia and ElevenLabs voices via the LiveKit Inference TTS integration and have enabled TTS‑aligned transcriptions. Note : I had tried the TimedString, but it is not able to import in the code.

Have you tried this?

tts=cartesia.TTS(  
            word_timestamps=True,  # This is default  
            # ... other settings  
        )

I am using livekit inference.