search

IAHispano / Applio

VITS-based Voice Conversion focused on simplicity, quality and performance.
https://applio.org
MIT License
1.36k stars 230 forks source link

[WIP] Different array lengths in v2 44k configs #428

Closed Tps-F closed 1 month ago

Tps-F commented 2 months ago

TL;DR

44k json config maybe broken. It looks like v1 architecture. Fix PR: https://github.com/IAHispano/Applio/pull/427

Codes

https://github.com/IAHispano/Applio/blob/d55124378974e7756282659270c94912c4ce39f6/rvc/configs/v2/44000.json#L43-L45

in L43 and L45, I think four array lengths is correct.

https://github.com/IAHispano/Applio/blob/d58f8262148d91364bfeb1b0fc2b80578ce8d834/rvc/lib/infer_pack/models.py#L259

and that may makes the length here wrong.

https://github.com/IAHispano/Applio/blob/d58f8262148d91364bfeb1b0fc2b80578ce8d834/rvc/lib/infer_pack/models.py#L266-L277

Convolutional Neural Network is also broken.

https://github.com/IAHispano/Applio/blob/d58f8262148d91364bfeb1b0fc2b80578ce8d834/rvc/lib/infer_pack/models.py#L687-L698

These generators will also be affected.