Abstract Speaker recognition is a biometric technique to recognize people’s identity based on their voice signal. A recognition system has two main requirements, which are high accuracy recognition rate and short processing time under large amount of training data. This paper propose a method to solve the two above requirements by performing a combination of two advantages of each VQ and GMM model to provide a new model can be called a “Hybrid VQ/GMM-UBM model”. This model not only takes the advantage of high accuracy in GMM method but also improve the accuracy rate and reduce the amount of computation of the system when combined with VQ model. The efficiency of the model is evaluated by computational time and accuracy rate compared to GMM models. Experimental results showed that the hybrid VQ/GMM-UBM model had better accuracy. Keywords Vietnamese Speaker recognition, Gausisian Mixture Model, Universal Background Model, Vector Quantization, Biometrics.