In UBM-GMM base speaker recognition, we need background GMM (UBM), how much data is enough, and what about adapting variances and weights to the desired speaker?? Or is it enough to adapat mean? 

More Rizwan Ishaq's questions See All
Similar questions and discussions