الفهرس Only 14 pages are availabe for public view
 |  | from | 124 |  |  |
| | | from | 124 | | |
Abstract
This thesis proposes two approaches building robust text-independent speaker identification systems in noisy environments. The first approach is based using robust statistical estimators such as the Minimum Volume Ellipsoid and Minimum Covariance Determinant to improve the maximum likelihood based classifiers by making them robust to outliers. The second approach is based on using an improved version for the large-margin discriminative criterion, which is characterized by its simplicity compared to the standard large-margin approach and other discriminative approaches. Experimental results show that our proposed techniques outperform the baseline techniques