| As an important branch of speech signal processing, speaker recognition has broad application prospects. The purpose of the present thesis is designing a speaker recognition system based on VI technology, which is developed for text-independent speaker recognition.From the principle speaker recognition is composed of speech signal acquisition, processing, the training of language model and identification, etc. The main work includes feature extraction of speech signal and selection and optimization of algorithm for building signal model. We have chosen mel-frequency cepstral coefficient as the feature parameter after studies on the various methods of feature extraction of speech signal. MFCC is extracted from preprocessing of speech signal on MATLAB. We use traditional VQ and first order HMM algorithm. Besides, a new kind of modified algorithm in VQ has been proposed .It is achieved by adding small feature samples codebook in speaker recognition.The system is developed on PC computer, powerful data processing of MATLAB and LabVIEW software. LabVIEW has many advantages, on which the virtual instrument system for speaker recognition is built. The accurateness could reach to 90% in VQ and could reach to 95% in HMM or more by the system experiment. This system has a lot of advantages, accessibility and friendly interfaces, etc. It satisfies the various needs of signal acquisition and real-time processing basically. |
