FAQ

Please make sure that you are using the latest version of the library.

Why training is so slow on my device?

• Please make sure that you are not using the CPU.

• If you are training a laptop, you may want to use FP32 instead of FP16.

Why the generated audio is blurry or weird?

• If it sounds noisy, you probably need to wait for more training steps.

• If it sounds blurry, please make sure you preprocessed the dataset using your current config.

Why I see KeyError ‘pytorch-lightning_version’?

This indicates that you are using a converted model from DiffSVC, which is not PyTorch Lightning. Please use --pretrained instead of --resume to load the model.

Why I see some error about missing keys when resuming training?

To support the ONNX exporting, we have to change the model structure. To solve this problem, you can use --pretrained instead of --resume to load the model.