This Space supports voice_clone only and is designed for Hugging Face ZeroGPU.

The runtime keeps device="auto" and does not expose frontend controls for switching CPU execution or backend backbone selection.

Upload a reference audio file or use a built-in preset voice, then generate the full output audio after decoding finishes.

The UI also supports selecting built-in example cases from assets/demo.jsonl, which auto-fill the text and preview the effective prompt audio.