search

MingjieChen / DYGANVC

demo page https://MingjieChen.github.io/dygan-vc
67 stars 9 forks source link

Why not train like StarGAN #13

Closed ywh-my closed 1 year ago

ywh-my commented 1 year ago

StarGAN training way could keep a many-to-many VC, why do you only use identity loss while lacking of consistent loss? And , is identity loss necessary in StarGAN training? Thank you .

MingjieChen commented 1 year ago

Hello,

I tried to use cycle-consistency loss as StarGAN, actually it made worse performance.

ywh-my commented 1 year ago

Dear chen, I would like to get your help。I use the losses like your DYGAN VC , including G_adv loss, identity loss , to train a emotional conversion framework. (using GST as emotional encoder , and I have a generator and discriminator . )

But I found that the convergence model could only reconstruct the input , but This model does not have the ability to change styles. (The output mel looks like the same as the input mel)

I do not know how to get a model that has style transfer ability?

MingjieChen commented 1 year ago

Hello,

This model is specifically developed for changing speaker identity. I don't think it will work for emotional transfer.

ywh-my commented 1 year ago

Thank you