On the use of classifiers for text-independent speaker identification

摘要

In this paper we have presented the comparative study of different modelling techniques (classifiers) for the text independent speaker identification. Four classifiers, namely, Gaussian mixture models, Fuzzy min-max neural network, Self organizing map, and Vector Quantization based Probabilistic Neural Network (VQ-PNN) have been used for the study. The database containing speech utterances recorded from forty two speakers in Hindi language was used for experimentation. Mel frequency cepstral coefficients that represent short time spectrum are used as features for identification. The performance of four classifiers is analysed under clean- and noisy-speech environment for different signal to noise ratios. All the four classifiers have almost similar performance for 10 second test speech utterances under clean environment. However, GMM outperforms other three classifiers under noisy test conditions.