Robust speaker identification based on hybrid model of VQ and GMM-UBM

摘要

In practical speaker identification applications, the performance of systems is generally degraded because of the presence of background noise. In this paper, an advanced hybrid model VQ/GMM-UBM in speaker identification which is a combination of Vector Quantization (VQ) and Gaussian Mixture Model-Universal Background Model (GMM-UBM) is presented. Even though this algorithm takes advantages of both VQ and GMM based identification algorithms, its identification efficiency, however, is also decreased in noisy environments. The impact of different types of noise on robustness of VQ/GMM-UBM based speaker identification is also evaluated in this paper. The experimental results indicated which types of noise speaker identification using VQ/GMM-UBM is robust against. According to the result, we can propose further that in practical applications affected by specific types of noise, whether or not noise reduction algorithms should be used to enhance VQ/GMM-UBM based speaker identification.