auspicious3000 / autovc

AutoVC: Zero-Shot Voice Style Transfer with Only Autoencoder Loss
https://arxiv.org/abs/1905.05879
MIT License
983 stars 207 forks source link

why the quality of the demo page has difference with your new paper? #36

Closed azraelkuan closed 4 years ago

azraelkuan commented 4 years ago

i listen all the demo wavs in these pages.

this is the new paper: https://auspicious3000.github.io/icassp_2020_demo/ this paper: https://auspicious3000.github.io/autovc-demo/

auspicious3000 commented 4 years ago

The new paper improved the autovc by controlling the pitch. The old paper focuses on timbre conversion, but the new paper introduces pitch conversion.

azraelkuan commented 4 years ago

@auspicious3000 what i mean is that the demo in [https://auspicious3000.github.io/autovc-demo/ ] is much better than the demo[autovc] in https://auspicious3000.github.io/icassp_2020_demo/

auspicious3000 commented 4 years ago

In order to show the improvements of the new method, we intentionally picked some problematic samples from the old paper.