Research on Automatic Speaker Recognition Based on Speech Clustering
Limin Xu, Bo Qian, Weiming Cheng, Zhenmin Tang · 2006
A novel approach of extracting speaker features based on speech clustering technology which employs both speaker's personality and speech cues is proposed in this work for speaker recognition. Specifically, the speech clustering-based approach is first applied before speaker feature extraction and multi-feature which is extracted respectively from the distinct clusters based on speech clustering is combined as a speaker multi-feature. Those two information cues including speaker's personality and speech are then effectively integrated using the multi-feature for achieving more robust results. The experiments show that using the template matching method the recognition rate with the multi-feature achieve 95.83%, which is 6.48% increase than the recognition accuracy with the long-term average feature. Moreover the average of false acceptance with the multi-feature decreased than the other ordinary features