基于多维参数的语音身份认证系统研究

The study on speaker recognition system based on Multi-Dimensional parameters

本文研究了与文本相关说话人语音通过多维参数的语音身份认证系统进行身份认证的过程。重点研究了小波降噪、语音增强、辅助加权MFCC语音特征提取,混合HMM语音辨识的算法及实现问题。试验表明该技术在认证效率、准确度、自适应性方面有较好表现。

A speaker recognition system to judge the speaker＇s identity according to his text-dependent voice based on Multi-Dimensional parameters is researched and realized in this paper. Some designs, such as using second generation wavelet to denoise, enhance speech and estimate the parametric distribution in speaker＇s spectrogram, extracting feature vectors with weighted MFCC and recognizing speaker＇s identity by mixed HMM algorithm on the appointed text, were, adopted to complete, the design. The experiments demonstrated that the improved technologies, including optimizing recognition ＇algorithm based on text composed of nasal, Modeling and implementing anisotropic orthorhombic wavelet base and mixed HMM algorithm, constructing a parallel Viterbi accelerator for block cyclic and convolution codes all had good performance in the aspects of identifying accuracy, running adaptation as well as robustness of the system.