Fusion of speaker and lexical information for topic segmentation: A co-segmentation approach

In this work, we investigate how speaker-based information and lexical-based information can be fused efficiently for topic segmentation of spoken contents. While in recent work, we have proposed an early fusion scheme, so as to jointly model speaker and lexical distribution, we propose here a co-segmentation framework, between segmentations performed in the speaker space and in the lexical space. Experiments carried out on two distinct corpora (Radio talk show and TV Broadcast News) show that, even if performances of speaker information are contrasted and closely related to the content structure, its integration with lexical information, either by early fusion or by co-segmentation, is always effective. Absolute gains of 16% (Radio corpus) and 5% (TV corpus) are observed for topic boundary detection performance.