Efficient implementation of GMM based speaker verification using sorted Gaussian mixture model

In this paper a new structured Gaussian mixture model, called sorted GMM, is proposed as an efficient method to implement GMM-based speaker verification systems; such as Gaussian mixture model universal background model (GMM-UBM) scheme. The proposed method uses a sorted GMM which facilitate partial search and has lower computational complexity and less memory requirement compared to the well-known tree-structured GMM of the same model order. Experimental results show that a speaker verification system based on the proposed method outperforms that of a similar system which uses tree-structured from performance point of view. It also provides comparable performance with the GMM-UBM method despite its 3.5 times lower computational cost for a GMM of order 64.