Research On Pathologic Voice Recognition Of Asymmetric Viscous Triangular Glottis Model

Voice is an important carrier of information exchange and emotional transmission.Due to environmental and occupational factors,the prevalence of voice diseases presents an obvious rising trend,which greatly affects people's health and quality of life.The research of pathologic voice recognition focuses on the acoustic detection combining acoustic feature extraction and pattern recognition.However,the relationship between vocal feature parameters and the actual physiological and pathological structure of vocal cords is lacking.Researching the vibration of vocal fold model can help us to understand the physiological characteristics and pathological mechanism of vocal folds,which is of great significance to reveal the characteristics of vocal cords vibration and the diagnosis of vocal fold diseases.Based on the analysis of the glottis structure and airflow distribution characteristics,an asymmetric viscous triangular glottis model is provided to reproduce the vibration pattern of diseased vocal cords.Firstly,the triangular glottis structure was constructed according to the characteristics of vocal cord structure.The state of glottis jet in glottis closure is analyzed and the pressure of asymmetric viscous airflow on the vocal cord wall is quantified by glottis inclination.The improved genetic algorithm which combines bidirectional neighborhood search and elite retention strategy is used to extract the feature parameters of the model.The correlation and difference of feature parameters of asymmetric viscous triangular glottis model are analyzed.The recognition of normal and pathological voices and the classification and recognition of pathological voices are carried out.Specific research contents are as follows:(1)The traditional three-mass block model ignores the difference of glottis structure and the limitation of glottis jet deflection in the process of vocal cord vibration.An asymmetric viscous triangular glottis model is proposed.To realize incomplete glottis closure when the left and right masses collide,a triangular glottis structure is constructed.The distribution of air flow at the glottis is obtained by the change of glottis inclination,so as to realize the effect of asymmetric air pressure on the vocal cord wall caused by the deflection of glottis jet.The effects of asymmetric coefficient,glottis pressure and the pre-phonation half-width of the glottis on the model output are analyzed.The simulation results show that the asymmetric viscous triangular glottis model is more sensitive to the change of the pre-phonation half-width of the glottis.The vibration asymmetry is more obvious under the pathological sound condition,and it can better display the characteristics of respiratory sound.(2)In order to avoid premature convergence of model feature optimization algorithm,on the basis of traditional genetic algorithm,elite retention strategy is introduced and bidirectional neighborhood search is carried out.The results of parameter inversion show that the weighted average error of the asymmetric viscous triangular glottis model is lower than that of the traditional three-mass block model,and the relative error of each glottis feature parameter is less than 2.5%.(3)The feature parameters of the model presented are used to assist pathological voice recognition.Furthermore,vocal cord paralysis,vocal cord polyp and vocal cord edema were identified by dichotomy.After the addition of vocal cord model features,the recognition rate of normal voice and pathological voice is increased by more than 10%,and the classification and recognition rate of pathological voice is up to 25%.The pathologic voice recognition effect associated with vocal cord edema is significantly improved.Finally,the shortcomings of this paper and the future research are pointed out.