Integration of mel-frequency cepstral coefficients with log energy and temporal derivatives for text-independent speaker identification

Abstract

This paper presents effect of possible integrations of delta derivatives and log energy with MFCC for text-independent speaker identification. MFCC features extracted from speech signal are used to create speaker model using vector quantization. First, the effect of varying MFCC filters and centroids of vector quantization is compared. Next, MFCC scheme is combined with delta derivatives and log energy. The effect of these possible combinations is compared by varying MFCC filters and centroids of vector quantization. Among all experiments carried out on 120 speakers of TIMIT database, average identification rate of 99.58 % is achieved for 29 MFCC filters and 32 centroids of vector quantization.