Bad result when infer with pretrained model. env problem?

KoeAI / LLVC

MIT License

372 stars 31 forks source link

Bad result when infer with pretrained model. env problem? #4

Closed Jackiexiao closed 10 months ago

Jackiexiao commented 10 months ago

python infer.py \
-p ./llvc_models/models/checkpoints/llvc/G_500000.pth \
-c ./experiments/llvc/config.json \
-f raw/ \
-o ori_output/ \
--stream -n 4

the output audio is noise....

orignial and convert audio file see:

https://transfer.sh/%28/rX9RCObKH5/libritts-103_1241_000054_000007.wav,/53x4gJo4jV/output-103_1241_000054_000007.wav%29.tar.gz

env

Python 3.10.13 torch.version '2.0.1'

Jackiexiao commented 10 months ago

I think it's my env's problem

Jackiexiao commented 10 months ago

A reproduce result on chinese aishell3 dataset, the audio quality is not good

https://transfer.sh/%28/mpJySjxf5d/girl-llvc-convert.wav,/OSaDcYBYf9/girl.wav%29.tar.gz