Detecting pathological speech using adaptive chaos exponent

This paper proposes an algorithm to detect pathological speech using adaptive exponent of chaos. Within the proper rang of sampling delay, the optimal Correlation Dimension (CD) parameters is searched with the aim of obtaining the minimum Equal Error Rate (EER) without pre-setting optional sampling delay or the range of optional embedded CD. In the saturation of CD curve, the CD curve is split into sub-curves consisting of five successive elements, and the differences between any two adjacent elements in the sub-curve are calculated. Then, the subset possessing the minimum differences is regarded as the stable sub-curve of the CD curves, and the third element of the stable subset is regarded as the optimal CD. Finally, after the EER analysis of the training data, the CD and its corresponding sampling delay which possesses the minimum EER are chosen as the parameters of chaos. The experimental results show that the proposed algorithm possesses the Classification Correct Rate (CCR) of 75.3%. Compared with the Shimmer algorithm, the Jitter algorithm, the SHR algorithm and FZA algorithm, 9%, 20.1%, 19.1% and 15.5% of improvements in CCR are respectively obtained.