Any Text Speaker Recognition System

Automatic speaker recognition is the processing of automatically recognizing who is speaking by using speaker specific information included in speech signal. With the development of communication and information technology, it is getting more and more attention for its bright future. It can be classified into speaker identification and speaker verification according to decision modes. This thesis focuses attention on free-text speaker identification. The main works are as follows: 1. Two speech corpuses, which include 15 speakers and 20 speakers respectively, are built. Some factors such as speed, volume and time interval which affect the performance of speaker identification system are taken into consideration. 2. The properties and extraction methods of some common feature parameters are studied in detail. These features include pitch period, FF1-based cepstrum, LPC-based cepstrum, CMS cepstrum, PFCMS cepstrum, PFL cepstrum and delta-cepstrum. 3. Four most popular and useful speaker models, vector quantization (VQ) model, Gaussian mixture model (GMM), group vector quantization (GVQ) model and radial basis function( RBF) network model, are studied respectively. Based on the above feature parameters and speaker models, some complete speaker identification systems are established. 4. Some methods, such as processing of features, combination of different features, mixed training, improvenent of decision rule, time scale modification of speech signals and adaptive speaker model, are studied in detail to improve the robustness of the speaker identification system. 5. The detailed testing results are given.