auspicious3000 / SpeechSplit

Unsupervised Speech Decomposition Via Triple Information Bottleneck
http://arxiv.org/abs/2004.11284
MIT License
636 stars 92 forks source link

mel spectrogram normalization range #51

Open inconnu11 opened 3 years ago

inconnu11 commented 3 years ago

Hi, I observed that the range of spectrogram saved in npy file is -0.2 ~ 0.8. I am wondering why you normalize spectrogram into this range? For what reason?

auspicious3000 commented 3 years ago

The spectrogram should be between 0 and 1. Anyway, the fast vocoder is released. See README.