A discrete wavelet based approach to speaker identification for different speaking styles

Abstract

This paper presents a speaker identification system using speech signals of different speaking styles. Normal and fast speaking styles have been chosen for the analysis. The speech data provided by the CHAINS corpus has been used for the experiment. The speech utterance used for testing is compared with the speech sample in the database. Two set ups, each consisting of three groups of speakers have been used for the experiment. From each of the setup, the information contained in the frequency ranges 300-4 KHz, 300-6 KHz and300-8KHz are extracted using the Daubechies db8 wavelet and the speaker identification accuracy is compared.