Sample rate and mono/stereo in configuration about transformertts

[pizzabug]

Hi! create-dataset.py uses librosa to import the wavs and automatically resamples the wav files to sampling_rate in preprocessing.

I've left it at 22050 because the MelGAN vocoder is fixed to 22050 hz, but sampling_rate doesn't have to be changed to reflect the dataset. I hope that helps.

[cfrancesco]

@pizzabug is exactly right, if you want the model to be usable with the pretrained vocoders leave it at 22050, otherwise you can change it to 16000 or anything that you find works well.

I am not sure regarding the stereo-mono question, in principle as long as it is possible to extract power spectrograms there should be no issue within the TTS model.