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

Does anyone know why parameters of C8C1I increases? #19

Open
TrilSiaWalt opened this issue Aug 17, 2024 · 3 comments
Open

Does anyone know why parameters of C8C1I increases? #19

TrilSiaWalt opened this issue Aug 17, 2024 · 3 comments

Comments

@TrilSiaWalt
Copy link

Snipaste_2024-08-17_22-06-50
Does anyone know why parameters of C8C1I increases?
I try to mainly modify the config as

    "upsample_rates": [8, 1],
    "upsample_kernel_sizes": [16, 15],

or

    "upsample_rates": [8, 1],
    "upsample_kernel_sizes": [16, 1],

but neither of their parameters changed as shown in the table.

@SynthAether
Copy link

inspecting what you are trying to do, it should be more like:

"upsample_rates": [8,1], "upsample_kernel_sizes": [16,2],

though that alone won't be enough, you also need to update the iSTFT parameters gen_istft_n_fft and gen_istft_hop_size so you get perfect upsampling to 256 samples per frame

@TrilSiaWalt
Copy link
Author

inspecting what you are trying to do, it should be more like:

"upsample_rates": [8,1], "upsample_kernel_sizes": [16,2],

though that alone won't be enough, you also need to update the iSTFT parameters gen_istft_n_fft and gen_istft_hop_size so you get perfect upsampling to 256 samples per frame

had done that. you can see that C8I also need it but it doesn't increase the parameters.

@NorthhhFishSoup
Copy link

they give reason in iSTFTNet2 paper:

Param of iSTFTNet-C8C1I32 is larger than that of iSTFTNetC8C8I4 because the number of channels is halved in the second 1D ResBlocks in iSTFTNet-C8C8I4, whereas it is not conducted in iSTFTNet-C8C1I32 owing to the absence of temporal upsampling. We used this strategy to confirm whether iSTFTNet-C8C1I32 could not obtain comparable speech quality, even with expressive modules.

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

3 participants