search

r9y9 / tacotron_pytorch

PyTorch implementation of Tacotron speech synthesis model.
http://nbviewer.jupyter.org/github/r9y9/tacotron_pytorch/blob/master/notebooks/Test%20Tacotron.ipynb
Other
306 stars 79 forks source link

Double denormalization? #6

Closed qbx2 closed 6 years ago

qbx2 commented 6 years ago

[1] https://github.com/r9y9/tacotron_pytorch/blob/master/synthesis.py#L57 [2] https://github.com/r9y9/tacotron/blob/5ec6822d823096fa4fd0f3e8a3b8eb639c164271/util/audio.py#L35 Here the output is denormalized at [1], and the spectrogram is denormalized at [2]. Is there any reason, or just a mistake?

r9y9 commented 6 years ago

It's correct. In the code, waveform is computed from normalized linear_output, not denormalized spectrogram. spectrogram is computed just for visualization used in jupyter notebook.

qbx2 commented 6 years ago

Aha. I see you're right. Thank you for explanation.