Voiceprint Recognition Technology Research

Voice print recognition is a sub-area of the general speech recognition problem. In contrast to the traditional speech recognition techniques which mainly aim to seek the semantic information contained in the sound signal, voice print recognition tries to identify the speakers directly from the sound signal.In voice print recognition, the production of the sound signal is usually modeled by the discrete time-domain model. With the respect to feature extraction, the requirement is that feature parameters selected should reflect both the physiological properties of the speaker's vocal organs and the psychological properties of the speaker's speech habits. Experiments show that a relatively good recognition performance could be achieved by combing such features as 1) the baseband and coefficients of inverse spectrum obtained using LPC analysis, 2) the difference-value of basis frequencies and coefficients of the difference-value of the inverse spectrum , which worked as the linear regressive coefficients. In summary, there exists various voice print recognition techniques with different advantages and disadvantages, and the choice of voice print recognition method should be task-specific. In this thesis, great attention has been paid on the problem of voice print recognition using RBF networks.