Light OmniVoice Demo

State-of-the-art text-to-speech model for 600+ languages, supporting:

  • Voice Clone — Clone any voice from a reference audio
  • Voice Design — Create custom voices with speaker attributes

Built with drbaph/OmniVoice-bf16 which is lighter than original model from Xiaomi AI Lab Next-gen Kaldi team but with the same quality.

Recommended: 3–10 seconds audio.

Language (optional) / 语种 (可选)

Keep as Auto to auto-detect the language.