Using the NPY output with a different vocoder? about multilingual_text_to_speech

[Tomiinek]

Hello,
do you have any specific vocoder in mind?

You can of course use arbitrary algorithm capable of inverting mel spectrograms (i.e., vocoders such as WORLD are not suitable and WaveGrad, MelGAN, WaveNorm, WaveGlow etc. are ok).
The output files are (hopefully) standard log-magnitude mel spectrograms of shape [num_mels, time].

[michael-conrad]

I see references to MelGAN, WaveGAN, WaveGlow, HiFi-GAN, etc. And would like to compare outputs with a couple of the newer ones.

[Tomiinek]

Ok and what is the question or problem? 😄
I think there shouldn't be any problem with using these vocoders.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.