Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

multi-track multi-domain genre transfer with stargan #6

Open
AllenPeng0209 opened this issue Nov 14, 2018 · 3 comments
Open

multi-track multi-domain genre transfer with stargan #6

AllenPeng0209 opened this issue Nov 14, 2018 · 3 comments

Comments

@AllenPeng0209
Copy link

Hi,
I'm trying to revise you code to do some multi-track multi-domain genre transfer with stargan, there are three track as my channel which are drum, bass and guitar. Here is some problem happened:

  1. It is easy to transfer drum pattern from one genre to another. However, it's really hard to transfer guitar and bass track. the problem would be like chord variation is almost unlimited, and it's relatively easier to transfer drum pattern.
  2. even though lambda is set to be 10 (cycle consistency loss parameter), the output of cycle_midi still can't not transfer B->A back in guitar and bass.

Do you have any idea to solve this problem?

your work is really interesting, thanks a lot.

@sumuzhao
Copy link
Owner

Hey, basically in my project, I just merge all the tracks into one single track. Multiple tracks would be way more difficult than single track. This is also what we can extend in this model. I suggest you could first use two tracks, one is drum and the other is guitar or just merge all other tracks. Regarding the starGAN, you'd better also try single merged track first then use multiple tracks. it's easier to control a single variant. Normally, the model should work well in a single track.

Currently, I don't have time to do any further work on this project. But feel free to contact me if you have any further questions. Good luck!

@AllenPeng0209
Copy link
Author

Thanks a lots for your answer, it's a good way to do more things on this model.

@AllenPeng0209
Copy link
Author

Some questions want to ask.
I found that the discriminator is always powerful than generator in my multi-track model, is there any parameter to adjust to solve this problem?
Thanks

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants