A New Codebook Design Method Based On Genetic Algorithm For Text-Independent Speaker Identification

Speaker recognition as one of the biometrics techniques is to recognize speaker's identity from its voice which contains physiological and behavioral characteristics specific to each individual. Speaker recognition has caught many attentions for its particularly advantage on convenience, economy and veracity and become an important and popular authentication technique in human life and work. Therefore, a more robust method for speaker recognition with high accuracy of recognition rate is the aim for researchers at home and abroad. In many speaker recognitions, paper combines VQ with GA, it uses Euclid distance to identify according to MFCC and its extended parameters.It often uses AMFCC to analyze change of every dimension parameter and AAMFCC to explain acceleration in the feature extraction. From old experiments and theories, paper introduces square sum of first-order derivative mel-cepstrum coefficient and difference which express total change of many-dimension cepstrum parameters in some frame. Along that method it appends two new feature parameters that are square sum of second-order derivative mel-cepstrum coefficient and its difference. Through experiments verified, new parameters increases the identification rate.Paper adopts the combination of VQ and GA, to avoid the possibility that classical LBG algorithm easily obtain local optimization and make up limitation of selection of initial codebook for final result. To prevent GA from getting into local optimization system not only increases iterative times, but also use a method that is combined by selecting part of optimal individuals and judging similarity every individual in the colony to change aberrance rate, it in some extent may avoid obtaining sub-optimization.From experiments, paper knows that after adopting GA recognition performance is improved comparing with LBG. And more long training speech length is, better recognition performance is.Paper finds that recognition does not always performance well while number of codeword is increasing. When the number reaches some, recognition begins to performances bad.