Research Into Speaker Recognition Technology And Its Application In Hospital Guide Platform

In this thesis, the speaker recognition technology and its application in Hospital Guide Platform is thoroughly discussed. As a convenient and practical biometrics identity verification technology, prevailing speaker recognition systems can obtain very high accuracy for clean speech, but their performance will degrade rapidly in noisy environments owing to the mismatch between the acoustic models and the testing speech. Therefore, noise robust technology is a crucial problem for the application of speaker recognition system in the hospital noisy condition.This dissertation first analyzes the main sources and categories of the noise in hospital, then we separate the noise into three classes according the constitution of noise in the different period, and build the speaker recognition database within the noisy environment. In view of current existence different speech modeling method, we compare many kinds of classified models through the experimental method. Through contrasting the recognition quality and the robustness between each kind of model, finally we choose the GMM-nv model as the basic model of the system.To decrease the mismatch between signal space, feature space and model space, we need to study and compare several fusion schemes. In feature space, we study some Spectral Variability Compensation techniques; in model space, we propose direct cepstral coefficients weighting GMM model based on different period, and weight every dimension of MFCC cepstral coefficients according to the respective discriminative ability during the process of recognition. In addition, we propose the maximize recognition criterion and united segment recognition criterion to use in the hospital environment recognition, enhancing the system recognition rate.In order to guarantee the recognition in each period has a high recognition rate, we adopt different fusion schemes in three different periods-the Work Period, the Break Period and the Mid-night Period. Through the research into the fusion schemes and the contrast reference experiment, we determine the final form of the recognition system. The results indicate that the method researched in this article can significantly increase the recognition accuracy in hospital noisy environments.