🎙️ OmniVoice Free Studio

Zero-shot voice cloning via the public k2-fsa/OmniVoice Space. Uses a short reference clip (3–10 s recommended).

Text to synthesize

Reference voice audio

.wav / .mp3 / .flac — ideally 3–10 seconds of clean speech

Reference transcript (optional)

Language

“Auto” is fine for most cases

Advanced (optional)

Instruct (voice design hint)

Inference steps

Guidance scale (CFG)

Speed

Duration (sec, 0 = auto)

Denoise Preprocess prompt Postprocess output

Generated audio