eDIAR removes fixed speaker-slot limits without retraining diarizers
Original title: eDIAR: Training-Free Open-Speaker Streaming Diarization for Long Meetings and Large Archives via e-means
Clustering with e-means starts only after every slot is used and keeps per-update cost from growing with meeting length.
eDIAR is a training-free method that lets fixed-slot streaming diarizers handle more speakers than their output slots. The slot head still decides who is active and when speech overlaps; speaker embeddings are clustered with e-means inside a slot only after every slot has been used. In streaming, re-clustering stays in a bounded window, labels are revised for a few seconds, and identities remain anchored to the head's slots. The same recipe converts four-slot Streaming Sortformer into an open-speaker system and is also applied to linking speakers across large archives.
- Streaming Sortformer and Nemotron-3 cannot name more speakers than their slots.
- On hour-long meetings the method stays far more accurate than Streaming Sortformer.
- e-means reaches Lloyd's k-means clustering quality several times faster.
- e-means uses its default recipe from a few hundred to more than a million embeddings.
From the original article
1 min read at huggingface.coMeetings last up to an hour, and nobody knows in advance how many people will speak. Neural diarizers with a fixed number of output slots, such as Streaming Sortformer and Nemotron-3, are accurate and run online, but they cannot name more speakers than they have slots. We remove this limit without retraining.
By Kyudan Jung, Jihwan Kim, Minwoo Lee · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.