Identification of Noisy Speech Signals using Bispectrum-based 2D- MFCC and Its Optimization through Genetic Algorithm as a Feature Extraction Subsystem

Power-spectrum-based Mel-Frequency Cepstrum Coefficients (MFCC) is usually used as a feature extractor in a speaker identification system. This one-dimensional feature extraction subsystem, however, shows low recognition rates for identifying utterance speech signals under harsh noise conditions. In this paper, we have developed a speaker identification system based on Bispectrum data that is more robust to the addition of Gaussian noise. As one-dimensional MFCC method could not be directly used to process the twodimensional Bispectrum data, we proposed a two-dimensional MFCC method and its optimization using Genetic Algorithm (GA). Experiments using the two-dimensional MFCC method as the feature extractor and a Hidden Markov Model as the pattern classifier on utterance speeches contained with various levels of Gaussian noise are conducted. Results showed that the developed system performed higher recognition rates compare with that of 1D-MFCC method, especially when the 2D-MFCC with GA optimization method is utilized. Key-Words: Speaker Identification System, 2D Mel-Frequency Cepstrum Coefficients, Bispectrum, Hidden Markov Model, Genetics Algorithms.