Sequence-Level Speaker Change Detection With Difference-Based Continuous Integrate-and-Fire

Speaker change detection is an important task in multi-party interactions such as meetings and conversations. In this paper, we address the speaker from perspective of sequence transduction. Specifically, propose a novel encoder-decoder framework that directly converts input feature to identity sequence. The difference-based continuous integrate-and-fire mechanism designed support framework. It detects changes by integrating difference between encoder outputs frame-by-frame transfers segment-level embeddings according detected changes. whole supervised sequence, weaker label than precise points. experiments on AMI DIHARD-I corpora show our sequence-level method consistently outperforms strong frame-level baseline uses labels.