Analysis And Recognition Of Pathological Speech In Patients With Dysarthria

Medical treatment and education for people with disabilities have always been highly valued in China.In recent years,scientific research has gradually turned to help groups in need,and the study of pathological phonetics in patients with dysarthria has received extensive attention.Previous studies on pathological pronunciation of Mandarin were mainly focused on acoustic analysis,and few studies were carried out in combination with kinematics.In this paper,the audio data and three-dimensional motion track data of patients with dysarthria and normal people were collected with the help of three-dimensional electromagnetic articulator,and the pronunciation data set of patients with dysarthria and normal people was established.The kinematics signal data collected simultaneously with acoustic signal data were analyzed.Aiming at the movement of tongue,lips and jaw of major articulation organs,the differences between articulation of patients with articulation disorder and normal people were explored.Based on the mechanism of articulation,the pathologic speech recognition and evaluation of patients with dysarsaras were studied theoretically and practically.The aim of this paper is to judge and evaluate the pronunciation problems of patients with dysarthria comprehensively and objectively,and to provide effective technical support and help for the medical pathological research and rehabilitation training of patients with dysarthria.This study has broad application prospects in medical and educational fields,so it has important theoretical significance and practical value.The main work and innovative results of this paper are as follows:(1)The three-dimensional electromagnetic phonograph is used to collect and establish the Chinese Mandarin phonetics data set,which covers the consonants,vowels,syllables and sentences of Chinese pronunciation.The data contains information about synchronous acoustics and kinematics characteristics of patients with dysarthria and normal subjects.(2)The acoustic and kinematic features of Mandarin phonemes from patients with dysarthria and normal subjects were extracted.The acoustic features included traditional acoustic features and nonlinear dynamic features.Kinematics features include movement displacement,velocity and the spatiotemporal index of pronunciation organs.The differences of acoustic and kinematical characteristics between patients with dysarthria and normal subjects were analyzed and compared to explore the pronunciation problems of patients with dysarthria.The simulation experiments of pathological speech recognition are also done for different acoustic and kinematic features.(3)Based on the characteristics of human ear hearing and nonlinear energy,a method for extracting the characteristic parameters of cochlear cepstrum coefficient(Cochlear Filter Cepstral Coefficients,CFCC)based on S-transform is proposed.The method combines not only the advantages of Fourier transform and wavelet transform,but also simulates the auditory perceptual characteristics of human ears from the perspective of bionics.The new feature is applied to pathological speech recognition and compared with the traditional features,the validity of the feature is proved.(4)Based on the sound-sound mechanism and the physiological characteristics of the sound-generating organ,a new characteristic parameter(Articulator Onset Time,AOT),which is a combination of acoustics and kinematics,is proposed,that is the starting time of the pronunciation movement.The results showed that the difference of the AOT parameter in the patients with dysarthria and the normal human was significant,and the judgment of the pathological speech had good regional division.The paper also makes a further study of the correlation between the acoustic and kinematic parameters.(5)A new fusion feature parameter algorithm based on kernel principal component analysis(Kernel Principal Component Analysis,KPCA)and discriminant canonical correlation analysis(Discriminative Canonical Correlation Analysis,DCCA)is proposed.The algorithm can reduce the correlation between the feature parameters and retain the important principal components in the feature parameters.The fusion feature vector is used for pathological speech recognition.Compared with the traditional combination features,the recognition rate is improved greatly.(6)A fuzzy comprehensive evaluation model for articulation quality of patients with dysarthria was established.The fuzzy set of acoustic and kinematic characteristic parameters is established in this model,and the objective weight coefficients of each parameter index are determined by F-Score algorithm.Combined with subjective expert evaluation,the comprehensive evaluation of pathological speech was carried out.This method can solve the problem of fuzzy and difficult to quantify,and make the comprehensive evaluation of pathological speech more comprehensive and objective.