25launches

The first launch of Reverb-TTS.

Reverb-TTS page

Reverb-TTS

Launching today

Local voice cloning, over 20x faster than real time on a Mac

Hunted by IlyasHunter
Visit huggingface.co (opens in a new tab)

Reverb-TTS is a text-to-speech model from Software Mansion AI Labs that clones a voice from a few seconds of speech and its transcript, with no fine-tuning. It speaks English, German, French, Spanish, Italian, Polish and Portuguese. At 297M parameters, pruned from OmniVoice, it runs over 20x faster than real time on an M4 Pro Mac or an iPhone 17, with int8, int4 and on-device builds. A browser demo runs on WebGPU. Code is Apache-2.0, but the weights are CC BY-NC 4.0: no commercial use.

Meet the team

Hunted by Ilyas. The makers are not on 25launches yet.

H (Hunter)
  • Reverb-TTS screenshot 1
  • Reverb-TTS screenshot 2
  • Reverb-TTS screenshot 3

Discussion 1

Ask questions, give feedback. Comments don't count in the ranking.

Sign in to join the discussion.

  • IlyasHunter Pinned

    Good voice cloning usually means a cloud API and your voice samples on someone else's server. Reverb runs on a laptop or a phone, and Software Mansion reports the lowest word error rate among the open models it compared on the Open TTS Leaderboard's MiniMax set. Check the license before building on it: the weights are for non-commercial use only.