Tight Boundary Prediction in Speaker Diarization Using Causal-Anticausal Consistency
For researchers and practitioners in speaker diarization, this method enables tight boundary prediction from loose labels, reducing annotation effort while improving downstream task performance.
The paper addresses the problem of loose speaker diarization boundaries caused by training on conversational data with margin-inclusive labels. It proposes a method using causal and anticausal models to generate tighter pseudo labels, recovering about 70% of the tightening effect of ideal tight-label training and improving downstream performance.
Multi-talker conversational automatic speech recognition data are often used to train speaker diarization models. Because such data prioritize semantic continuity, pauses and boundary margins are included within speech segments, resulting in loose annotations. Models trained on such data tend to internalize mechanisms that reproduce this looseness, although tight speech intervals are sometimes preferable for downstream applications. In this paper, we address the novel task of enabling models to produce tight predictions using loose labels. Our method generates tighter pseudo labels using causal and anticausal models, which are inherently incapable of learning loosening behavior. We further propose a co-training scheme that iteratively tightens labels and updates both models for more progressive refinement. Experimental results show that the proposed method recovers about 70 % of the tightening effect achieved by ideal tight-label training and improves downstream performance.