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

About the variable 'input_audio_corrupt' in lines 240-241, isn't it target_audio_corrupt? #27

Open
taehoGit opened this issue Sep 21, 2022 · 1 comment

Comments

@taehoGit
Copy link

# ------------------------ Target audio ----------------------
# use the same augmented audio clip, add different random EQ and compressor
target_audio_corrupt = input_audio_aug.clone()
# apply frequency and dynamic range corrpution (expander)
if self.freq_corrupt and torch.rand(1).sum() < 0.75:
target_audio_corrupt = augmentations.frequency_corruption(
[target_audio_corrupt], self.sample_rate
)[0]
# peak normalize again before passing through dynamic range compressor
input_audio_corrupt /= input_audio_corrupt.abs().max()
input_audio_corrupt *= 10 ** (-12.0 / 20) # with min 3 dBFS headroom
if self.drc_corrupt and torch.rand(1).sum() < 0.75:
target_audio_corrupt = augmentations.dynamic_range_compression(
[target_audio_corrupt], self.sample_rate
)[0]

@csteinmetz1
Copy link
Collaborator

Hi @taehoGit, this does look like a bug, however it likely has a minimal effect. There may be some different behavior between the dynamics of the input and the target, but style transfer training should still function. Thanks for pointing it out.

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