Skip to content

Less aggressive SpecAug to reduce(?) divergence#2548

Merged
asumagic merged 1 commit intospeechbrain:developfrom
asumagic:conformer-training-fix
May 16, 2024
Merged

Less aggressive SpecAug to reduce(?) divergence#2548
asumagic merged 1 commit intospeechbrain:developfrom
asumagic:conformer-training-fix

Conversation

@asumagic
Copy link
Copy Markdown
Collaborator

@asumagic asumagic commented May 13, 2024 •
edited
Loading

What does this PR do?

Fixes #2533 (potentially; not verified).

This is a reduction of ~20% of the freq-wise and the time-wise SpecAugment, because we found out those seem too aggressive and would make it so that some samples had very few features unmasked.

This is a WIP that hasn't been tested extensively.

TODO:

  • Check whether the final model has a similar (or better) WER.
  • Check whether the model training actually becomes reliable (need to test over a bunch of trains, usually visible at ~1.5 epochs).
Before submitting
  • Did you read the contributor guideline?
  • Did you make sure your PR does only one thing, instead of bundling different changes together?
  • Did you make sure to update the documentation with your changes? (if necessary)
  • Did you write any new necessary tests? (not for typos and docs)
  • Did you verify new and existing tests pass locally with your changes?
  • Did you list all the breaking changes introduced by this pull request?
  • Does your code adhere to project-specific code style and conventions?

PR review

Reviewer checklist
  • Is this pull request ready for review? (if not, please submit in draft mode)
  • Check that all items from Before submitting are resolved
  • Make sure the title is self-explanatory and the description concisely explains the PR
  • Add labels and milestones (and optionally projects) to the PR so it can be classified
  • Confirm that the changes adhere to compatibility requirements (e.g., Python version, platform)
  • Review the self-review checklist to ensure the code is ready for review

@TParcollet TParcollet marked this pull request as ready for review May 15, 2024 10:10
TParcollet
TParcollet approved these changes May 15, 2024
View reviewed changes
Copy link
Copy Markdown
Collaborator

@TParcollet TParcollet left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM. We will need to wait and see...

@asumagic
Copy link
Copy Markdown
Collaborator Author

asumagic commented May 16, 2024

Final WER is very close after 100 epochs, merging. Will reopen the issue if it still arises.

@asumagic asumagic merged commit dde2311 into speechbrain:develop May 16, 2024
@asumagic asumagic deleted the conformer-training-fix branch October 22, 2024 15:00
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@TParcollet TParcollet TParcollet approved these changes

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

Training regression for Conformer-Transducer models

2 participants

@asumagic @TParcollet