Preprint
DIHARD II is Still Hard: Experimental Results and Discussions from the DKU-LENOVO Team
Qingjian Lin et al., arXiv, 2020
Read the originalRead on October 17, 2026
- What it is
- A team report on its submission to the second DIHARD speaker diarization challenge. It lists the modules of the system and gives the team's error rates on the two challenge tracks.
- How much weight it carries
- A team's own report of its system, not the official challenge leaderboard. Numbers apply to the challenge's hard, mixed-domain test data, which is harder than most meeting recordings.
Facts taken from this source
- The DKU-LENOVO diarization system is built from several separate modules, including voice activity detection, segmentation, speaker embedding extraction, clustering, resegmentation and overlap detection.
“Our diarization system includes multiple modules, namely voice activity detection (VAD), segmentation, speaker embedding extraction, similarity scoring, clustering, resegmentation and overlap detection.” (Abstract)
- In the second DIHARD challenge, the DKU-LENOVO team's diarization system reached a diarization error rate of 18.84 percent on Track 1.
“Our proposed system achieves 18.84% DER in Track1 and 27.90% DER in Track2.” (Abstract)
- In the second DIHARD challenge, the DKU-LENOVO team reached a diarization error rate of 27.90 percent on Track 2, from the same system that reached 18.84 percent on Track 1.
“Our proposed system achieves 18.84% DER in Track1 and 27.90% DER in Track2.” (Abstract)