Research of the speaker verification based on the SVM-GMM mixture model

摘要

We put forward a new SVM-GMM mixture model to improve recognition rate of the speaker verification system in the paper. Support vector machines (SVM) and Gaussian mixture model (GMM) are widely applied to the speaker verification, but both have some disadvantages. We present a new approach for speaker verification based on their feature. The new model introduce the output of the Gaussian mixture model to Support vector machines, in order to adjust the probabilistic output of the support vector of machines.It can compliment support vector machines with probabilistic information. The experiments have proved that SVM-GMM mixture model can effective enhance the recognition rate of the speaker verification system.