Speaker Recognition Technology With Limited Vocabulary

The speaker recognition technology analyses the input voice, extracts the speaker’s feature information and recognizes the identity automatically by using the information. As a kind of biometrics technology, the speaker recognition technology can solve critical private information. At the same time, it is the simpleness of implement, convenience of operation and easyness of input voice to make the speaker recognition technology applying to many practical products.Due to the highlighted aging problem, the old industry is required to be developed. We should research and design an old fitted product to solve the problems encountered by old people. Applying the speaker recognition technology to this kind of product should improve the safety further.Through studying the traditional feature extraction algorithm and the pattern recognition algorithm, this thesis propose an MFCC algorithm based on entropy weighting and an initialized EM algorithm based on spectral clustering. To study the performance of the algorithm, this thesis also set up a new speech database. The last part of this thesis designs a kind of portable devices used by old which applied the speaker recognition technology to improve the safety. Besides, a software test platform for the handheld devices has also been constructed.