how can an change an agent's voice or speed during the call (elevenlabs tts)?

Thanks for reaching out, Francisco. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

what about node js?

Thanks — you're asking how to change an agent’s ElevenLabs TTS voice or speaking speed during a call in Node.js. You can update the TTS options at runtime using the session TTS instance’s

update_options

method; changes take effect on the next utterance. Node.js example

// assuming you're using the ElevenLabs TTS plugin and have access to the agent session
class MyAgent extends voice.Agent {
  async switchVoiceAndSpeed() {
    // Update voice ID and adjust speed for next utterance
    this.session.tts?.update_options({
      voice_id: 'EXAVITQu4vr4xnSDxMaL', // new ElevenLabs voice ID
      speed: 1.2,                      // 1.0 = normal, 0.5-2.0 supported range
      stability: 0.8,                  // optional ElevenLabs voice params
      similarity_boost: 0.9            // optional
    });
  }
}

Key points • Use

voice_id

to select the ElevenLabs voice. • Use

speed

(e.g., 0.5–2.0) to slow down or speed up speech; values >1.0 increase speed. • Other ElevenLabs parameters (like

stability

,

similarity_boost

,

style

) can also be provided. • Changes apply to subsequent utterances — they do not interrupt the current speech. If you want, tell me: • The voice_id you plan to use, or the kind of voice (male/female, tone) you want, and • Whether the change should be one-off (next utterance) or persistent for the session. I can provide a complete runnable snippet showing how to trigger this from an event or tool in your agent flow. Sources: ElevenLabs TTS plugin guide | LiveKit Documentation

what are the options to access the agent session? the other participant updating its localParticipant parameters?

Thanks — I’m responding to your question about how to access the AgentSession (so you can update ElevenLabs TTS voice/speed) and whether changing another participant’s

localParticipant

parameters is a valid approach. Short answer • Access the AgentSession directly (recommended): from inside an Agent class (

this.session

/

self.session

), from the RTC session entrypoint before/after

session.start()

, or from a tool lifecycle via the

RunContext

(

context.session

). • No — updating another participant’s

localParticipant

parameters is not a valid way to change the agent’s TTS settings. Use the session’s TTS API instead (e.g.,

session.tts.update_options()

). How to access AgentSession (options) • From an Agent subclass ◦ Inside your agent logic you can reference the session directly:

this.session

(Node) or

self.session

(Python). ◦ This is the most direct place to call

this.session.tts.update_options(...)

. • From the RTC session entrypoint ◦ The AgentSession is created in the rtc_session entrypoint; you can keep a reference when you instantiate it and use it before/after

session.start()

. • From a function tool or lifecycle hook ◦ Tools receive a

RunContext

(or similar) that exposes

context.session

, which is the current AgentSession. • From global job context ◦ The framework exposes job-level context (AsyncLocalStorage). You can access the job context to reach session-related objects when appropriate. Why not use

localParticipant

updates •

localParticipant

is for room-level participant state (publishing tracks, metadata/attributes) and is meant for client/frontend synchronization. • The AgentSession controls the AI pipeline (STT, LLM, TTS). Updating participant metadata will not change the agent’s internal TTS configuration. • To change voice or speed reliably, call the TTS API on the session:

session.tts.update_options({ voice_id: "...", speed: 1.2 })

. Changes take effect on the next utterance. If you want, I can: • Show a concrete Node.js example that updates

this.session.tts.update_options()

from a function tool or event handler. • Help you decide where to trigger the update in your call flow (e.g., on a user command, on a specific intent, or on a UI action), and provide the exact snippet for that case. Which would help most: a short runnable snippet inside an Agent method, or an example showing how to trigger it from the frontend via metadata/command? Sources: Agent session | LiveKit Documentation | Build a multilingual voice agent that automatically switches languages | Li... | Tool definition and use | LiveKit Documentation | Agents (Node.js) | livekit/livekit_composite | DeepWiki | Pipeline nodes and hooks | LiveKit Documentation