Fact · Transcription
The DKU-LENOVO diarization system is built from several separate modules, including voice activity detection, segmentation, speaker embedding extraction, clustering, resegmentation and overlap detection.
Checked against its source on October 17, 2026
In the source's own words
“Our diarization system includes multiple modules, namely voice activity detection (VAD), segmentation, speaker embedding extraction, similarity scoring, clustering, resegmentation and overlap detection.”
DIHARD II is Still Hard: Experimental Results and Discussions from the DKU-LENOVO Team, Qingjian Lin et al., arXiv, 2020, Abstract.
Open the original- What it measures
- Each module can introduce its own errors: a missed speech segment, a wrong speaker cluster, or an overlap that is not split.
- What it does not show
- Describes one team's system. Other systems use end-to-end models that merge these steps.