ISCA Archive - Speaker recognition based on feature space trace (original) (raw)

Speaker recognition based on feature space trace

Wu Yadong, Li Zhizhu

This paper presents a multiple templates matching algorithm based on feature space trace, which is used in speaker recognition. It extracts the cepstrum coefficient as feature parameter. We normalize the sequence of feature parameter based on feature space trace. The fuzzy c-means method is adopted in generating the multiple templates and the multiple matching method is applied to match the templates. Experiments show this method is a robust, high recognizable and valuable way to implement text-dependent speaker recognition.