Recurrence Quantification Analysis And Cation Of Pathological Voice

Voice is the most primitive and convenient way of human information interaction.With the development of technology and economy,it shows its superiority in intelligent human-machine interaction.However,in real life,due to the improper use of voice and other factors,the vocal folds are affected,thus affecting normal speech communication and physical and mental health.The vibration of pathological vocal folds is complex.Although the traditional acoustic characteristics have their own acoustic characteristics,they can not directly reveal the non-linear characteristics of pathological voice.The traditional non-linear characteristics(correlation dimension and largest Lyapunov exponents)can not precisely analyze the characteristics of different pathological voice.To the issues discussed above,this paper proposes a recurrence threshold adaptive algorithm for pathological voice from the perspective of nonlinear dynamics recurrence,solves the recurrence threshold to construct a recurrence plot to analyze the recurrence characteristics of pathological voice intuitively,and compares the recurrence characteristics of vocal fold paralysis,vocal fold edema,vocal fold polyp and vocal fold nodule;in order to improve the ability of pathological differentiation to classify the above four kinds of pathological voice,a recurrence threshold adaptive multi-hierarchy recurrence quantification analysis method is designed.In addition,in order to save the time cost of pathological voice recognition,the principal singular spectrum of concept recurrence matrix is used to quantify the complexity.The main contents are as follows:(1)The typical threshold selection method needs subjective experience.Improper selection will lead to the situation that some recurrence structures in the recurrence plot cannot appear or false recurrence occurs.Aiming at the problem that the typical recurrence thresholdselection method can not be applied to pathological voice of different types and pathological changes in varying degrees,a recurrence threshold adaptive algorithm is proposed.Through the observation and analysis of pathological voice recurrence plots under three different recurrence thresholds,it is found that the recurrence plots obtained by this method can better reveal the recurrence characteristics of pathological voice,and make the above four recurrence quantification measures of pathological voice distinctive.In order to reveal more recurrence features to represent the pathological features of voice,a recurrence threshold adaptive multi-hierarchy recurrence quantification method is designed by introducing hierarchical decomposition.After the multi-hierarchy decomposition of the signal,the corresponding adaptive recurrence threshold is solved,and then the recurrence quantization measure is solved,and the statistical difference analysis is carried out.According to the two-hierarchy recurrence plot,the similarities and differences of the recurrence structures of vocal fold paralysis,vocal fold edema,vocal nodule and vocal fold polyp were analyzed.By using the strategy of OvO decomposition,the problem of four kinds of pathological voice classification is transformed into six sub problems of two classification,and the results of six base classifiers are aggregated by the strategy of distance weighting.The accuracy of the above four kinds of pathological voice classification in MEEI database is 73.15%.(2)For pathological voice recognition,concept recurrence map is introduced to reduce the computational complexity of dynamic recurrence map.The accuracy of pathological voice recognition in MEEI database is 99.58%by using the quantitative measure of complexity of concept recurrence matrix-principal singular spectrum ratio,combined with the texture characteristics of concept recurrence matrix.Finally,it points out the shortcomings of this paper and the future work direction.