Add DER with details of overlapped error#71
Draft
FrenchKrab wants to merge 4 commits into
Draft
Conversation
hbredin
reviewed
May 9, 2025
| ) | ||
| comps_ovl = self.der_ovl.compute_components( | ||
| reference, hypothesis, uem=ovl_regions | ||
| ) |
Member
There was a problem hiding this comment.
Not sure I understand the logic here.
compute_components can actually lead to different mapping whether we focus on overlap or non overlap regions (because they have their own internal call to optimal_mapping). There is no garantee that this is the same mapping.
Or did I miss something?
Author
There was a problem hiding this comment.
No, that's correct ... it was supposed to be an IdentificationErrorRate. But the tests passed because there was no overlap in the hypothesis. I think now it should be correct.
Fixed it with the last two commits.
(the test passed with an incorrect metric because the hypothesis contained no overlap)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This PR adds a variant of the DiarizationErrorRate that splits every error component in two: overlapped and non-overlapped.
For example,
der['false alarm'] == ovlder['ovl false alarm'] + ovlder['nonovl false alarm'].This allows better diagnosis of where diarization errors are coming from. Might be useful since overlapped speech is very tricky :)
The code for the metric should be done (and seems to work well, even with different collar sizes according to the tests). I still need to add code for plotting this metric.