论文部分内容阅读
提出了一种嗓音多频带非线性分析的声带病变识别方法,以提高声带病变嗓音的识别率。首先采用Gammatone听觉滤波器组对嗓音信号进行滤波,求取每个频带下的最大李雅普诺夫指数;对映射到核空间的数据采用高斯最大似然度准则优化核函数,然后采用优化核主成分分析算法实现特征抽取。识别实验表明,多频带最大李雅普诺夫指数的识别率比传统的MFCC和最大李雅普诺夫指数分别有6.52%和8.45%的提高,且采用优化核主成分分析算法比传统核主成分分析算法有更好的抽取效果.将多频带非线性分析和优化核主成分分析算法结合,识别率提升至97.82%。
A vocal cord lesion identification method based on multi-band non-linear analysis of speech was proposed to improve the vocal cord recognition rate. Firstly, the Gammatone auditory filter is used to filter the voice signal to get the maximum Lyapunov exponent in each frequency band. The kernel function is optimized by Gaussian maximum likelihood criterion for the data mapped to the kernel space, and then the optimized kernel principal component Analysis algorithm to achieve feature extraction. The recognition experiments show that the recognition rate of multi-band maximum Lyapunov exponents is 6.52% and 8.45% higher than that of traditional MFCC and maximum Lyapunov exponents, respectively. Compared with the traditional principal component analysis algorithm Better extraction effect.Combining multi-band nonlinear analysis with optimized kernel principal component analysis, the recognition rate is improved to 97.82%.