Why lws and not regular stft?

[duvtedudug]

Is there a reason for using lws?

Do you use reconstructed phase somewhere to improve quality?

Thanks in advance, Duvte.

[r9y9]

Do you use reconstructed phase somewhere to improve quality?

No.

The reason is that I wanted to integrate the vocoder with https://github.com/r9y9/deepvoice3_pytorch, which uses lws. I needed to use same audio feature extraction pipeline. See r9y9/deepvoice3_pytorch#21 for wavenet vocoder integration with DeepVoice3.

[duvtedudug]

That makes sense, thanks!

How close are you to integrating DeepVoice3 together with WaveNet Vocoder?

[r9y9]

It will take me a few months :(

[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.