Specialgfhdhdh/singing-voice-cloner-rvc

Singing Voice Cloner (RVC v2)

Fine-tuned RVC v2 model for singing voice conversion.

Training Details

• Base: Seed-VC (Plachta/Seed-VC) - zero-shot voice conversion
• Architecture: Diffusion Transformer (DiT) with F0 conditioning
• Singing Model: 44kHz, 128 mel bins, 17 layers, 12 heads
• F0 Extractor: RMVPE
• Vocoder: BigVGAN v2 44kHz

Components

• Vocal Separation: Demucs htdemucs (SOTA source separation)
• Voice Cloning: Seed-VC F0-conditioned model (zero-shot, 44kHz)
• TTS: Edge-TTS (Microsoft Neural TTS)
• Mixing: pydub audio overlay

Usage

See the Singing Voice Cloner Space for the full pipeline.

Pipeline

1. Upload a song → Demucs separates vocals + instrumental
2. Enter new lyrics → Edge-TTS generates speech
3. Seed-VC converts TTS speech to match the singing voice (F0-conditioned)
4. Mix new vocals with instrumental → final output

Credits

• Seed-VC by Plachta
• Demucs by Facebook Research
• Edge-TTS

Generated by ML Intern

This model repository was generated by ML Intern, an agent for machine learning research and development on the Hugging Face Hub.

• Try ML Intern: https://smolagents-ml-intern.hf.space
• Source code: https://github.com/huggingface/ml-intern