VoiceFixer Models

Official VoiceFixer model checkpoints for speech restoration, mirrored for faster downloads.

Why This Repository?

The official VoiceFixer models are hosted on Zenodo, which can have slow download speeds. This repository mirrors the same official models on HuggingFace for:

Faster downloads via HuggingFace CDN
Better integration with HuggingFace-based projects
Improved reliability and availability

Models Included

vf.ckpt (466.6 MB) - Analysis module for speech restoration
model.ckpt-1490000_trimed.pt (129.3 MB) - Vocoder module (44.1kHz)

Usage

Used by TTS Audio Suite for ComfyUI audio restoration.

Attribution

Original models by Haohe Liu et al.

Paper: VoiceFixer: Toward General Speech Restoration
GitHub: https://github.com/haoheliu/voicefixer
Original Zenodo Release: https://zenodo.org/record/5600188

License

CC-BY-4.0 (same as original release)

These models are identical to the official Zenodo release, mirrored here for convenience.