Skip to main content

​
Voice providers

Sonara supports multiple text-to-speech (TTS) providers, each offering different voice styles, languages, and quality levels. Choose the provider and voice that best fits your agent’s personality and use case.

​
Available providers

​
Selecting a voice

1

Open voice settings

Navigate to your agent’s configuration and click the Voice tab.
2

Choose a provider

Select a TTS provider from the dropdown. Each provider offers different voices.
3

Browse and preview voices

Browse the available voices for your selected provider. Each voice shows:
  • Name
  • Gender
  • Accent/language
  • Preview button to hear a sample
Click the play button to preview a voice before selecting it.
4

Adjust speed

Use the speed slider to adjust how fast the agent speaks. The default is 1.0x. Lower values slow the speech down; higher values speed it up.

​
Sarvam voices (Bulbul v4 Flash)

Sarvam agents use the Bulbul v4 Flash model (bulbul:v4-flash). Voices are personas: the same speaker recorded in different languages and styles. Each persona ID has the form <name>_<language>_<style>, for example aparna_hi_customer or simran_en_customer.
  • The voice picker groups personas by speaker. Each chip shows the persona’s language and style.
  • Recommended for support calls: aparna_hi_customer (female) and shubh_hi_customer (male).
  • Sample rate: 8, 16, 22.05 or 24 kHz.
  • Speed: 0.5x to 2.0x.
  • Audio previews aren’t available for v4 voices yet.
Agents created before Bulbul v4 show their voice as legacy v3. They keep working with that voice until you pick a v4 voice and publish.

​
Language support

Sonara supports 60+ languages with automatic language detection for multi-language agents.

​
Single language mode

Select one language for your agent. The agent will only speak and understand this language.

​
Multi-language mode

Enable multi-language support to let your agent automatically detect and respond in the caller’s language. You can configure:
  • Supported languages - Which languages the agent can use
  • Fallback language - The default language if detection is uncertain
Not all voice-language combinations are supported. If your selected voice doesn’t support a language, Sonara will automatically suggest a compatible alternative.

​
Speech-to-text (STT)

The STT provider determines how your agent transcribes what callers say. Configure this in the voice settings:
  • Provider - Select an STT provider (Deepgram is the default)
  • Model - Choose the transcription model (e.g., nova-3 for highest accuracy)
Deepgram’s nova-3 model offers the best balance of speed and accuracy for real-time voice conversations.