So I have been trying to integrate Tavus realtime avatar for a project and there seems to a long delay before the agent starts speaking. If we look at the video feed, its black for around 13 sec after which the avatar video is really up and speaks. Would love to know if anyone else has faced similar issue.
13 seconds is very long - I gave this a test and see around half that with the following setup:
- GitHub - livekit-examples/agent-starter-python: A complete voice AI starter for LiveKit Agents with Python. modified to use the Tavus avatar and run with
uv run … dev(it would have been quicker if I had deployed the agent to LiveKit cloud rather than locally throughdev) - Front end: Agent playground
We also have a dedicated example for Tavus: https://github.com/livekit-examples/python-agents-examples/tree/main/complex-agents/avatars/tavus with its own front end
Got it…i was also playing around with other providers. Tried liveavatar with the same setup and did a latency benchmark wrt to the avatar silence
5 sessions :
Tavus - 11.56sec avg
Liveavatar - 2.56sec avg
I will also try the above tavus repo to verify things. Thanks @darryncampbell
Thanks, we see more and more requests for benchmarks (specifically for LLM, STT, TTS) but I’ll raise that avatar benchmarks would also be beneficial; I’m sure others are in the same position as you and looking for this info.