Use more than one voice

Create a team of agents.

Multi-voice allows you to assign multiple voices to an agent. Since each voice has distinct qualities, this setup can simulate a full team of agents handling conversations. You can configure this in any function, though this guide assumes you’re using the start function.

Currently, setup is only possible through code configuration, not through the Agent Studio UI.

Select your voices

To use multiple voices, you need their ElevenLabs voice IDs. Refer to the ElevenLabs documentation to find them.

Here is an example of how to configure multiple voices as part of the start function. This example is for an agent that will be randomly assigned a voice at the start of each interaction:

def start_function(conv: Conversation):
    conv.randomize_voice([
        VoiceWeighting(
            voice=ElevenLabsVoice(
                voice_id="LcfCDJNUPlGQjkzn1xUU",
                similarity_boost=0.2,
                stability=0.4
            ),
            weight=0.5
        ),
        VoiceWeighting(
            voice=ElevenLabsVoice(
                voice_id="SOYHLrjzKX21zeoPC6cr",
                similarity_boost=0.7,
                stability=0.3
            ),
            weight=0.5
        )
    ])

How it works

• conv.randomize_voice([...]): Selects a voice at random based on the assigned weights.
• VoiceWeighting: Associates a voice with a probability of being selected.
• ElevenLabsVoice: Specifies an ElevenLabs voice with configurable parameters:
  • voice_id: The unique identifier for the voice.
  • similarity_boost: A value (0.0 to 1.0) that controls how closely the generated voice matches the original.
  • stability: A value (0.0 to 1.0) that controls the consistency of tone and delivery.
  • For more information on these parameters, visit the ElevenLabs API documentation.
• Weights: Define selection probability:
  • First ElevenLabsVoice: 0.5 is 50%
  • Second ElevenLabsVoice: Also 50%, so they are equal.
  • The sum of all weights must equal 1.0.

Understanding voice weights

You can add a maximum of five voices to an AI agent.

The weight parameter determines how often each voice is selected. The sum of all weights must equal 1.0.

Examples:

• One voice at 1.0: Always uses that voice.
• Two voices at 0.5 each: Each is selected 50% of the time.
• Four voices at 0.25 each

Personalizing repeat caller experiences

Repeat callers are more likely to ask for a human agent if they hear the same AI voice every time. Switching voices on subsequent calls can make interactions feel less robotic, and it is proven to reduce immediate escalations.

Using multi-voice may prevent callers from instantly requesting an agent just because they recognize the AI.