Talking Avatar plugin for LiveKit Agents

Support for Talking Avatar realistic avatars: a photorealistic avatar that joins the room and speaks your agent's voice, with lip sync, from any LLM, TTS or realtime model.

See the Talking Avatar LiveKit docs for more information.

Installation

pip install livekit-plugins-talkingavatar

Pre-requisites

You'll need an API key from the Talking Avatar dashboard. It can be set as an environment variable: TALKING_AVATAR_API_KEY. The plugin also reads LIVEKIT_URL, LIVEKIT_API_KEY and LIVEKIT_API_SECRET to mint the avatar's token.

Usage

from livekit.agents import Agent, AgentSession, JobContext
from livekit.plugins import talkingavatar


async def entrypoint(ctx: JobContext):
    session = AgentSession(
        # stt=..., llm=..., tts=... or a realtime model
    )

    avatar = talkingavatar.AvatarSession(avatar_id="ava4")  # any avatar ID from your dashboard
    await avatar.start(session, room=ctx.room)

    await session.start(agent=Agent(instructions="You are a friendly assistant."), room=ctx.room)

The avatar joins the room as its own participant and publishes, on behalf of your agent:

Your agent no longer publishes its own audio. Interruptions stop the avatar mid-sentence, and the agent's transcript is cut at what the user actually heard.

Billing is $0.02 per minute, per second, from the moment the avatar is ready until the session ends.