Guides / Driving the avatar

Driving the avatar#

There are three ways to make the avatar speak, plus barge-in to stop it.

Through the conversational model#

send_message sends user text through the server's model. The model's reply is synthesized and spoken by the avatar, and streams back to you as events.

await session.send_message("What's the weather like on Mars?")

Optionally override the avatar for this turn:

await session.send_message("Say that again, but excited!", avatar="presenter-male-1080")

The reply arrives as MESSAGE_STREAM_EVENT_RECEIVED chunks followed by a final MESSAGE_RECEIVED, and is added to the conversation history.

Straight to speech#

talk bypasses the model and speaks your exact text via TTS. It does not affect conversation history.

await session.talk("This line is spoken immediately.")
await session.talk("With a little attitude.", tone="confident")

Incrementally with a talk stream#

When text arrives a bit at a time (e.g. tokens from your own model), stream it so the avatar starts speaking before the sentence is complete.

async with session.create_talk_stream() as talk:
    await talk.send("Streaming ")
    await talk.send("this ")
    await talk.send("out loud.", end_of_speech=True)

The first chunk is automatically flagged as the start of speech; mark the last one with end_of_speech=True (or just exit the async with, which closes the stream). See Talk streams for the full API.

Barge-in#

Stop the avatar mid-utterance at any time:

await session.interrupt()

This is the same barge-in the live voice pipeline uses — the avatar eases back to its idle/listening pose. A TALK_STREAM_INTERRUPTED event fires.

Choosing between them#