Sarvam Instant Voice Cloning
Sarvam's beta voice-cloning API can save a voice once or accept a reference clip with each generation. It supports one-shot, HTTP-streamed and WebSocket speech generation.
Preview
Quality control depends on the endpoint
The one-shot endpoint can transcribe generated speech and retry outputs that fail its quality check. The guide limits this check to short, single-chunk generations. It is not a blanket quality guarantee for streamed speech.Instant Voice Cloning API
Practical use
Practical guidance. Use saved voices when consistency across calls matters. Compare pronunciation and identity across target languages, with permission to use the reference voice. Evaluate streaming behavior separately from short one-shot clips.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Preview |
| Access | API |
| Input choice | Send either voice_id or ref_audio, not bothInstant Voice Cloning API |
Related models
- Bulbul V4 Flash
Sarvam AI ยท Bulbul
A text-to-speech model with a persona-based catalog of 222 voice IDs.
Launch date not confirmedAPI