Turn detection support for local languages (Hindi, Punjabi, Tamil)

Agents
1

This question originally came up in our Slack community and the thread has been consolidated here for long-term reference.

I’m building a voice agent for conversations in local languages like Hindi, Punjabi, and Tamil.

Questions:

  1. Is there a way to use turn_detection with local languages?
  2. If I have to use VAD, is there a way to pass the language in metadata and use turn_detection when it’s a supported language, and fall back to VAD when it’s not?
2

  1. The multilingual turn detection model will work with Hindi, but for Punjabi and Tamil you would have to use VAD.

  2. You can switch turn detection dynamically using session.update_options:

Enable multilingual turn detection:

session.update_options(turn_detection=MultilingualModel())

Fall back to VAD:

session.update_options(turn_detection="vad")

You would need to configure the language beforehand or determine it during the call (e.g., using LLM), as most STT providers don’t report the detected language on the fly.