Open Kamii-Sam opened 2 months ago
Did you use a pretrained model? If not, 3400 training steps is not enough for convergence and requires at least several hours of audio data
Yes, I did use a pretrained model. I managed to actually fix the issue by turning on pitch augmentation in the config settings.
I trained a model and with the output, some parts of it sound squeaky and other parts sound deeper than usual.
python main.py -i input.wav -m exp/reflowvae-wavenet-attention/model_3400.pt -o output.wav -k 0 -f 0 -tid 1 -step 50
Above is the command I'm using. I'm confused as to why it's doing this.