yoyolicoris / wavenet-like-vocoder

Basic wavenet and fftnet vocoder model.
19 stars 2 forks source link

hop_length questions #1

Open maozhiqiang opened 5 years ago

maozhiqiang commented 5 years ago

hi, Thank you for your hard work, why set hop_length to 128 in you config file, How does this parameter affect audio quality?

yoyolicoris commented 5 years ago

@maozhiqiang The two config files are just templates for these two model. I found these values work for the model so I left them unchanged. You can try increase the hop_lenght to 256 or 512, I think it shouldn't make a big difference.

maozhiqiang commented 5 years ago

@yoyololicon thanks! I will try it

rishikksh20 commented 5 years ago

@yoyololicon Do you have any voice samples generated from this repo ?

yoyolicoris commented 5 years ago

@rishikksh20 Here's a short piano clip generated from a wavenet using the same config file in the repo, trained on MAPS dataset after 300k steps. Haven't try on voice data before. test.zip