← 研踪 PaperTrailarXiv:2610.34707 · v1
S2SND-v2: Unified Online and Offline Speaker Diarization with Speaker-Adapted w2v-BERT 2.0
首次提交 2026-10-26 09:52
版本更新 2026-10-26 09:52
雷达观测 2026-10-29 00:36
摘要
This paper presents S2SND-v2, a unified online and offline speaker diarization framework built on S2SND that improves acoustic feature extraction and the use of the resulting representations. For acoustic feature extraction, we introduce a speaker-adapted w2v-BERT 2.0 front end and compare full and pruned construction routes. To better use these features, overlap-aware activity-query masking improves speaker embedding extraction under overlap, while late-speaker training aligns unknown-speaker training with online inference conditions. Combining these improvements, the pruned system achieves competitive average diarization error rates of 14.52% online and 12.03% offline across eight test sets.
作者备注：Submitted to ICASSP 2027. 5 pages, 2 figures, 2 tables
观测到的版本
相邻记录可能跳过中间版本；首次发现 v2 及以上也会标为版本更新。
v1 · 2026-10-26 09:52 · 最新观测版本
S2SND-v2: Unified Online and Offline Speaker Diarization with Speaker-Adapted w2v-BERT 2.0
This paper presents S2SND-v2, a unified online and offline speaker diarization framework built on S2SND that improves acoustic feature extraction and the use of the resulting representations. For acoustic feature extraction, we introduce a speaker-adapted w2v-BERT 2.0 front end and compare full and pruned construction routes. To better use these features, overlap-aware activity-query masking improves speaker embedding extraction under overlap, while late-speaker training aligns unknown-speaker training with online inference conditions. Combining these improvements, the pruned system achieves competitive average diarization error rates of 14.52% online and 12.03% offline across eight test sets.
查看此版本解读 →查看此版本原文 ↗