Recognition and study of pathological voice based on nonlinear dynamics using gaussian mixture model/support vector machine / 生物医学工程学杂志

ABSTRACT

In the traditional identification of pathological voice, linear analysis techniques are usually used to analyze the characteristics of voice, and the linear classical model is often considered to be approximate to of the real voice production process. However, this must have ignored the nonlinear characteristics of voice in the actual generation process. In the paper, based on the nonlinear dynamics analysis method, the pathological voice is analyzed quantitatively and 7-dimensional nonlinear features, Hurst parameter, time delay, the second-order Rényi entropy, Shannon entropy, correlation dimension, Kolmogorov entropy and the largest Lyapunov e exponent are extracted. The experimental results showed that the method of nonlinear dynamics could compensate the deficiencies of the traditional methods, and could analyze normal and pathological voice well. Gaussian mixture model (GMM) and support vector machine (SVM) methods for pattern recognition were used to discriminate the test set including 39 cases of normal and 36 cases of pathological voice, and a better recognition rate is received, 97.22% and 97.30% for each of the mentioned normal and pathological cases, respectively.