Genetics Algorithm for 2D-MFCC Filter Development in Speaker Identification System Using HMM

Abstract

In this paper, we introduced the using of Genetics Algorithm to optimize the construction of a 2D filter in 2D-MFCC technique for speaker identification system. The 2D filter construction was very essential for processing the bispectrum data which is represented in 2D space instead of the usual 1D power spectrum. By using the 2D filter, the conventions; ;~;dden· Markov Model could be used as ih~ speaker classifier for processing 2D bispectrum data that was robust to noisy environment. The experimental comparison with IDMFCC technique and 2D-MFCC without GA optimization shows that a comparable high recognition for original uttered voice. However, when the uttered voice buried in Gaussian noise at 20 dB, the developed 2D-MFCC shows in higher recognition of 88.5% whilst only 59.4% and 70.5% for 1D-MFCC and 2D-MFCC without GA, respectively.