search

p0p4k / vits2_pytorch

unofficial vits2-TTS implementation in pytorch
https://arxiv.org/abs/2307.16430
MIT License
478 stars 86 forks source link

model compute error #8

Closed pipijiev12 closed 1 year ago

pipijiev12 commented 1 year ago

RuntimeError: Given groups=1, weight of size [192, 80, 1], expected input[64, 513, 297] to have 80 channels, but got 513 channels instead After I preprocessed the data, everything went fine. However, model calculation problems arise during the training process. Is it because I preprocessed the data wrong?

p0p4k commented 1 year ago

@pipijiev12 Hi, it should be fixed in the latest patch. Thanks for letting me know! It is because we are using 80 channel mel spectrogram in this architecture instead of 513 channels linear spec that VITS-1 uses. I had a typo in data_utils.py, I believe the problem should be fixed now.