A histogram based speaker identification technique

Abstract

Feature extraction has the capability to improve the performance of speaker identification systems. This paper proposes two new techniques for speaker identification based on utilizing a reduced set of the features generated from the Mel Frequency Cepstral Coefficient method (MFCC). These techniques are based on histograms for the features using pre-defined interval lengths. The first technique builds a histogram for all data in the feature vectors for each speaker while the second technique builds a histogram for each feature column in the feature set of each speaker. Speaker identification is based on the Euclidian distance measure.