Advances in SVM-Based System Using GMM Super Vectors for Text-Independent Speaker Verification