Sarvam CreativeVoice Cloning (Beta)

Cloning Best Practices

View as Markdown

Recording Quality Comes First

If a clone doesn’t sound quite right, the fastest fix is almost always a cleaner re-recording rather than tweaking downstream settings - recording quality has more influence on the final persona than any setting you can adjust afterward.

  • Follow the recording screen’s own guidance: speak clearly, in a quiet room, with no music playing.
  • Read the provided passage naturally, at your normal pace - don’t rush just because it’s a short recording.
  • If the first take doesn’t feel right, use Not happy? Re-record during the Personalise step rather than finishing with a take you’re unsure about.
  • If a passage feels awkward to read aloud, use Shuffle to get a different one rather than stumbling through it - a clean read of an easy passage beats a shaky read of a hard one.

Choosing a Language

  • Record in the language you’ll actually use the clone for - the capture is language-specific, so a clone made in English won’t carry the same accent nuance if you deploy it primarily for Hindi content.
  • If you need the same persona across multiple languages, consider creating a separate clone per language rather than assuming one capture generalizes across all of them.

Timing and Testing

  • Clone a voice before you need it for a live project, not during - processing takes a little time.
  • Test the result against a few sample lines in Text to Speech before committing it to a real script, so a re-do doesn’t land in the middle of production.
  • Only clone voices you have explicit rights to use.
  • Name profiles specifically at creation time (see Managing Your Voice Library) so you’re not guessing which clone is which later.