same content and voice in testing

GANtastic3 / MaskCycleGAN-VC

Implementation of Kaneko et al.'s MaskCycleGAN-VC model for non-parallel voice conversion.

MIT License

111 stars 31 forks source link

same content and voice in testing #12

Open todalex opened 3 years ago

todalex commented 3 years ago

Hi everyone so I started testing after the debug @HikaruHotta did .and i reached at gloss 7.5 and dloss 0.29 so i decided to convert an output and see if the result is good . but there is a problem here .my output wav file is from speaker s1 to s2 but its the same evaluation file i have for s1 added some noise .its no VC its just the same file with some noise. can u tell me whats the problem?

ganjiid commented 3 years ago

@HikaruHotta yeah pls explain this one

hikaruhotta commented 3 years ago

@todalex could you please pull again. I've just approved a PR with some patches

todalex commented 3 years ago

i think u should changed the readme too some train arguments are not used now like lr