I am currently working on a conversational voice agent and would like to explore whether LiveKit supports (or plans to support) configurable backchannel behavior during active user speech.
Specifically, I am looking to implement functionality where the conversational agent can speak configurable backchannel words (e.g., “mm-hmm”, “I see”, “right”, etc.) while the user is actively speaking. The goal is to make the interaction feel more natural and reassure the user that they are being heard — without interrupting or cutting off their speech.
The key requirements are:
-
The agent should emit backchannel words only from a predefined configurable list.
-
The agent must not interrupt or override the user’s speech.
-
The timing between backchannel utterances should be controllable (e.g., randomized within a defined interval range).
-
No words outside the configured list should be spoken.
Could you please let me know:
-
Whether this type of functionality is currently supported?
-
If not natively supported, whether there are recommended patterns using LiveKit agent to implement this behavior?
I would appreciate any guidance or documentation you can share.