witcherxz/habibi-voice-studio

Habibi Voice Studio

A free ZeroGPU long-form demo for SWivid/Habibi-TTS, with reference-audio voice cloning and all 13 dialect IDs supported by the unified model. Article text is split into reference-duration-aware Arabic/Latin sentence batches. A fixed consistency seed, level matching, smooth crossfades, and peak headroom reduce voice drift across the final WAV. The public demo accepts up to 5,000 characters per request; the Docker API in the source project supports larger articles and batches.

Use only voice samples you have permission to clone. The unified model weights are licensed under CC-BY-NC-SA-4.0 and are restricted to non-commercial use.