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

some advice about MAS Algorithm #12

Closed
BridgetteSong opened this issue Jun 23, 2021 · 3 comments
Closed

some advice about MAS Algorithm #12

BridgetteSong opened this issue Jun 23, 2021 · 3 comments

Comments

@BridgetteSong
Copy link

As the GlowTTS MAS definition:
MAS

to satisfy monotonicity and surjection, if z_j and {u_i, σ_i} are aligned, the previous latent variable z_j-1 should have been aligned to
either {u_i-1, σ_i-1} or {u_i, σ_i}, which is equal to the Forward Attention for Tacotron2(https://arxiv.org/abs/1807.06736), my question is:

  1. replacing max(Q_i-1_j-1, Q_i_j-1) with log(exp(Q_i-1_j-1) + exp(Q_i_j-1)) is a better choice?
@BridgetteSong
Copy link
Author

I will do ablation experiments and post my results.

@BridgetteSong BridgetteSong changed the title about MAS Algorithm About MAS Algorithm Jul 9, 2021
@BridgetteSong BridgetteSong changed the title About MAS Algorithm some advice about MAS Algorithm Jul 9, 2021
@xiaoyangnihao
Copy link

xiaoyangnihao commented Feb 8, 2022

I will do ablation experiments and post my results.

@BridgetteSong Hi,how about your ablation experiments and results?

@BridgetteSong
Copy link
Author

BridgetteSong commented Feb 9, 2022

@ xiaoyangnihao

  1. there are some bugs in new MAS, I need time to fix bugs, but I don't much time to do this.
  2. I emailed to the author to discuss the MAS. Now I think a soft alignment is key to improve quality, and my method maybe only improves stability. But author's MAS is stable enough, so I don't experiment more.

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