Skip to content

Bug fix: oscillating cluster threshold in agglomerative-cluster#2333

Merged
danpovey merged 2 commits intokaldi-asr:masterfrom
mmaciej2:diarization-fix
Apr 3, 2018
Merged

Bug fix: oscillating cluster threshold in agglomerative-cluster#2333
danpovey merged 2 commits intokaldi-asr:masterfrom
mmaciej2:diarization-fix

Conversation

@mmaciej2
Copy link
Contributor

@mmaciej2 mmaciej2 commented Apr 3, 2018

There was a bug in agglomerative-cluster.cc that would cause the threshold used for clustering to oscillate between positive and negative per recording.

@danpovey
Copy link
Contributor

danpovey commented Apr 3, 2018

Merging. Please comment whether this would have affected any checked-in recipes.

@danpovey danpovey merged commit 6f4fafb into kaldi-asr:master Apr 3, 2018
@mmaciej2
Copy link
Contributor Author

mmaciej2 commented Apr 3, 2018

This should not affect any checked-in recipes, as they tune the threshold as part of the recipe. It's possible that the error rate will decrease slightly, but using PLDA scoring should result in an optimal threshold of 0 in theory; and, a threshold of 0 would be unaffected by this bug, so the effect should be minimal. I will verify that there is no appreciable performance difference on the CallHome recipe, but this bug should really only affect scoring methods that are not "centered" around 0, which we do not currently use.

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

Successfully merging this pull request may close these issues.

2 participants

Comments