Text-Independent Speaker Identification Using Average Spectrum And GMM Approaches

摘要

A text-independent speaker identification system is implemented in this paper. Average FFT spectrum feature is used as the first approach to identify different speakers, because of its relatively good performance and low cost. Then, Gaussian Mixture Model(GMM) approach is used to model the utterance more precisely and to reduce the noise affections. The solving algorithm of the two approaches are given, and some important factors affecting the efficiency and reliability of the system are focused on, such as the decision threshold, the preemphasis coefficient, GMM model order, utterance length for testing and training.