The pretrained checkpoints for VoiceFixer, mirrored from Zenodo, from which voicefixer downloads them on import, at about 1 MB/s. The filenames are the ones voicefixer expects.
vf.ckpt: the restoration network, from Zenodo record 5600188, md51f479dabb9fa5724de5e20f2164cbc7a.model.ckpt-1490000_trimed.pt: the TFGAN vocoder, from Zenodo record 5469951, md5ac6de16561dd6042daf8bb3a806dc818.
Both records are licensed CC-BY-4.0, by Haohe Liu. The voicefixer code is MIT.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support