Voice Selection & Voice Cloning

Your AI assistant can speak with built-in voices or a custom cloned voice. Natural, realistic voices increase customer trust and engagement.

​

TTS Provider

Language, provider, model and voice all live in Voice & speech → Language & voice, and they are picked in that order: the provider list depends on the language you chose, so the dropdown only appears once a language is set. Available in Pipeline and Dualplex modes. Providers:

• ElevenLabs - high-quality voices, a large library
• Cartesia - fast, low-latency synthesis
• Inworld - where enabled for your account; it does not support secondary languages
• Azure - available for a smaller set of languages

​

Models

Most providers offer more than one speech model, and the Models choice appears right under the provider when they do. It is the same voice either way — the model decides how fast it answers, how expressive it sounds, and which languages it can speak. Roughly, each provider offers a fast model and a more expressive one: A model that cannot speak one of your languages is shown as Not available in the selected language and cannot be picked. This is why a language sometimes leaves you only one model: it is the only one that speaks it. Providers with a single model, such as Azure, show no Models choice. Speech-to-speech does not have one either: there the model generates the voice itself, so it is chosen as the engine’s model rather than here.

​

Voice Library

Each TTS provider has its own voice library. Select male/female, accent, or language based on your provider.

​

Importing Voices from the Provider Library

Beyond the built-in catalog, you can import any public voice from your TTS provider’s own library (ElevenLabs Voice Library or Cartesia Library). Available in Pipeline and Dualplex modes. Steps:

1. Click “Import voice” next to the voice selector
2. Pick the provider (ElevenLabs or Cartesia)
3. Open the provider’s voice library and find a voice you like
4. Copy the voice link or ID (the ⋮ menu next to the voice)
5. Paste it in the import field and click Import

You’ll get a notification when the voice is ready — it then appears in the voice picker like any other voice, billed at the same flat per-minute rate. Notes:

• ElevenLabs voices with a custom price set by their creator cannot be imported.
• Private or unshared voices cannot be imported — only public library voices.

​

Voices in Speech-to-Speech

In Speech-to-speech there is no separate synthesizer: the multimodal model speaks for itself. You pick that model in General → Engine type, as Multimodal model, and the voice picker offers the voices that model can use — multilingual ones, so there is no language filter and no provider to choose. Your own cloned voices stay in the list whatever the assistant’s language is.

​

Voice Cloning

Clone a voice from an audio sample. Available in Pipeline and Dualplex modes, and limited by your plan — the picker tells you when you have used up your cloned voices. Clone to Provider:

• Cartesia - Single audio file, at least 10 seconds, 1 speaker, no background noise
• ElevenLabs - Samples over 1 minute, 1 speaker, no background noise. Max 5 minutes total. Quality over quantity.

Steps:

1. Click “Clone voice” next to voice selector
2. Select provider (Cartesia or ElevenLabs)
3. Choose the voice language
4. Enter a name for your voice
5. Record or upload audio
6. Wait for processing
7. Select your new voice from dropdown

​

Best Practices

1. High-Quality Audio: Clearer samples give better results
2. Steady Delivery: Natural tone, no abrupt changes
3. No Background Noise: Record in a quiet environment
4. Legal: Ensure permission to clone voices that aren’t yours

---

Tip: After selecting or cloning a voice, do a test call to confirm it sounds as expected.