OlaWod
Question about the scale of gaussian noise in the mel spectrogram
In the proposed SR-augmentation, when the mel is squeezed, you pad it with the highest frequency bin value and add Gaussian noise.
Can you share the scale of the added gaussian noise?
(I guess it can differ depending on how you preprocess the audio to get the mel. I am currently using the 'mel_spectrogram_torch' function of https://github.com/jaywalnut310/vits/blob/2e561ba58618d021b5b8323d3765880f7e0ecfdb/mel_processing.py
If you used a different code for mel processing, it would be great to know.)
Also, could you tell me if you had issues with Nan losses (when applying the SR augmentation)?
The scale of the added gaussian noise is 0.1. We use the same mel processing. We did not encounter Nan losses. The SR augmentation is conducted offline, i.e., we synthesize multiple augmented waveforms , extract their WavLM features, store them on disk, and load them as input to the model.
When there is a hole in the high frequency spectrum, I have experienced nan loss.
