A kernel trick for sequences applied to text-independent speaker verification systems