Vocos Versions Save

What's Changed

• New multi-resolution (+multi-band) discriminator adopted from Descript Audio Codec.
• Updated recommended hyperparameters for the AdamW optimizer: lr=5e-4, betas=(0.8, 0.9)
• Pre-trained models on Hugging Face have been updated: charactr/vocos-mel-24khz, charactr/vocos-encodec-24khz

:bulb: Note: If you'd like to load a previous checkpoint, they have been tagged for easy reference:

Vocos.from_pretrained("charactr/vocos-encodec-24khz", revision="v0.0.4")

Fix types errors https://github.com/charactr-platform/vocos/pull/10 thanks to @alealv

Resolves https://github.com/charactr-platform/vocos/issues/15, https://github.com/charactr-platform/vocos/issues/16

Fix inefficient phase unwrapping (https://github.com/charactr-platform/vocos/pull/7) thanks to @roebel

🐶 Bark integration

Initial release