Two microphones speech enhancement systems based on instrumental variable algorithm for speaker identification

In this paper, we study the performance limits of a standard GMM speaker identification (SI) system in "adverse conditions" context using several real noises. Adaptive noise cancellation represents one such potentially effective technique and refers to a class of adaptive enhancement algorithms based on the availability of a primary input source and a secondary reference source. It will be shown that with the use of an adaptive noise cancelling called the Double Generalized Least Mean Squares (DGLMS), the SI performance can approach the optimal performance system. Experiments are done on Spidre corpus corrupted by different type of noises. The performance of the SI is improved by more than 35%.