Skip to content

[WIP] implement power low compression loss #18

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.

Sign up for GitHub

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

Open
stegben wants to merge 1 commit into
base: master
Choose a base branch
from
Open

[WIP] implement power low compression loss #18

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions config/default.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,6 +39,7 @@ train:
final: 0.05
summary_interval: 1
checkpoint_interval: 1000
complex_loss_ratio: 0.1 # the lambda in power-law compression loss computation
---
log:
chkpt_dir: 'chkpt'
Expand Down
13 changes: 10 additions & 3 deletions utils/train.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -64,9 +64,16 @@ def train(args, pt_dir, chkpt_path, trainloader, testloader, writer, logger, hp,
mask = model(mixed_mag, dvec)
output = mixed_mag * mask

# output = torch.pow(torch.clamp(output, min=0.0), hp.audio.power)
# target_mag = torch.pow(torch.clamp(target_mag, min=0.0), hp.audio.power)
loss = criterion(output, target_mag)
# Power-law compression
magnitude_loss = criterion(
torch.pow(torch.abs(output), hp.audio.power),
torch.pow(torch.abs(target_mag), hp.audio.power),
)
complex_loss = criterion(
torch.pow(torch.clamp(output, min=0.0), hp.audio.power),
torch.pow(torch.clamp(target_mag, min=0.0), hp.audio.power),
)
loss = magnitude_loss + complex_loss * hp.train.complex_loss_ratio

optimizer.zero_grad()
loss.backward()
Expand Down