A COMPARATIVE-STUDY OF ROBUST LINEAR PREDICTIVE ANALYSIS-METHODS WITHAPPLICATIONS TO SPEAKER IDENTIFICATION

In this paper, various linear predictive (LP) analysis methods are stu died and compared from the points of view of robustness to noise and o f application to speaker identification. The key to the success of LP techniques is in separating the vocal tract information from the ptich information present in a speech signal even under noisy conditions. I n addition to considering the conventional, one-shot weighted least-sq uares methods, we propose three other approaches with the above point as a motivation. The first is an iterative approach that leads to the weighted least absolute value solution. The second is an extension of the one-shot least-sqaures approach and achieves an iterative update o f the weights. The update is a function of the residual and is based o n minimizing a Mahalanobis distance. Third, the weighted total least-s quares formulation is considered. A study of the deviations in the LP parameters is done when noise (white Guassian and impulsive) is added to the speech. It is revealed that the most robust method depends on t he type of noise. Closed-set speaker identification experiments with 2 0 speakers are conducted using a vector quantizer classifier trained o n clean speech. The relative performance of the various LP approaches depends on the type of speech material used for testing.