基于Feature warping和RASTA滤波的GMM-UBM说话人确认

In speaker recognition application, there is a need to extract information from speech that is speaker specific and robust to noise and various channel. Though the GMM-UBM (Gaussian Mixture Model-Universal Background Model) system has proven to be very effective for text-independent recognition tasks, when there exists channel mismatch, additive noise and nonlinear effects due to handset transducers, the short-term distribution of the speech features can be distorted by these adverse effects, the performance of the system degrades. In this paper, some efforts has been done to minimize the frailty, Feature warping has been used to compensate for channel variations and RASTA filters have also been used to alleviate the convolutional and additive noise. The Experiments show that this technique is a suitable method to enhance speaker recognition robustness under adverse conditions.