Speaker Verification Using Adapted Bounded Gaussian Mixture Model

In this paper, we propose the application of bounded Gaussian mixture model (BGMM) to speaker verification. In the proposed approach, BGMM is employed for universal background model (UBM) and adapted speaker model. The proposed UBM is a large BGMM trained to represent speaker-independent distribution of features. In adapted speaker approach, hypothesized speaker model is derived by adapting the parameters of BGMM based UBM using speaker's training speech and maximum a posteriori (MAP). We have applied TIMIT and TSP speech corpora for the development of UBM and further testing of speaker verification by adapted speaker model. The proposed framework has demonstrated its effectiveness by improved speaker detection rate.