A Weighted Measure of Similarity for Speaker Tracking

In this paper, we present a speaker tracking system entirely based on anchor models approach. This speaker tracking system is combined with a speaker clustering module which improves speaker detection performances. However, speaker clustering errors generate new speaker detection errors due to segments grouped into wrong clusters. The aim of this article is to introduce a weighted measure of similarity between target speakers and segments based on a measure of similarity between segments belonging to a same cluster. Evaluation is done on the audio database of the ESTER evaluation campaign for the rich transcription of French broadcast news. Results show that the weighted measure of similarity improved the speaker tracking performances. This improvement manifests itself as an improved precision rate on segments grouped into wrong clusters