Speaker Identification Using 2-d Dct, Walsh and Haar on Full and Block Spectrogram

This paper aims to provide different approaches to text dependent speaker identification using DCT, Walsh and Haar transform along with use of spectrograms. Spectrograms obtained from speech samples are used as image database for the study undertaken. This image database is then subjected to various transforms. Using Euclidean distance as measure of similarity, most appropriate speaker match is obtained and is declared as identified speaker. Each transform is applied to spectrograms in two different ways: on full image and on image blocks. In both the ways, effect of different number of coefficients of transformed image is observed. Haar transform on full image reduces multiplications required by DCT and Walsh by 28 times whereas applying Haar transform on image blocks requires 18 times less mathematical computations as compared to DCT and Walsh on image blocks. Transforms when applied to image blocks, yield better or equal identification rates with reduced computational complexity.