wonjune-kang / lvc-vc

End-to-End Zero-Shot Voice Conversion with Location-Variable Convolutions
https://arxiv.org/abs/2205.09784
MIT License
86 stars 6 forks source link

[Code] Preprocessing code for LVC-VC #4

Closed hcy71o closed 1 year ago

hcy71o commented 1 year ago

Thank you for your interesting research. I was wondering if you have any code related to [lifting in the Quefrency domain] mentioned in the paper. If I'm not mistaken, there seems to be only preprocessing code for LVC-VC-XL (wav2vec2.0). If it is not disclosed, I'd be grateful to know the source you referenced in your implementation for preprocessing LVC-VC.

wonjune-kang commented 1 year ago

Hello! The low-quefrency liftering is done directly in the dataloader in dataset.py rather than being part of the preprocessing pipeline because it has a fairly low computational cost. Please see the function lowquef_lifter in utils/utils.py for more details on the actual implementation.

hcy71o commented 1 year ago

Hello! The low-quefrency liftering is done directly in the dataloader in dataset.py rather than being part of the preprocessing pipeline because it has a fairly low computational cost. Please see the function lowquef_lifter in utils/utils.py for more details on the actual implementation.

Thank you for kind explanation!