Research On Pathological Voice Repair Algorithm Based On Nonlinear Model

In recent years,affected by factors such as living environment,personal habits,and occupational characteristics,the number of patients with voice diseases has increased dramatically.Because voice diseases can cause hoarseness and ambiguity,which seriously affects communication efficiency and quality of life,there is an urgent need to repair pathological voice processing technology.Repairing pathological voice through non-invasive speech signal processing technology can improve the clarity and intelligibility of pathological voice signals,and help patients with voice diseases experience more convenient voice interaction.Pathological voice repair technology has broad application prospects in the medical field.The pathological voice repair framework consists of three parts: excitation parameter repair,vocal tract parameter repair,and speech synthesis.Aiming at the problem of low intelligibility of pathological voice,this thesis uses the above repair framework to design a pathological voice repair system based on a nonlinear model,and uses suitable data set to objectively evaluate the system.The specific research work is mainly reflected in the following aspects:(1)In terms of excitation parameter repair,in order to solve the problem of inaccuracy and instability in the extraction of the pitch of pathological voices by traditional algorithms,pitch extraction algorithm suitable for pathological voice is proposed: Use energy separation algorithm to extract the amplitude envelope of pathological voice signal;Use spectral harmonics correlation and normalized cross correlation function to extract the pitch trajectory;Select the best pitch trajectory based on dynamic programming.(2)In terms of vocal tract parameter repair,the vocal tract characteristics of pathological voices are reconstructed by correcting the formant frequency and bandwidth.Aiming at the inaccuracy and instability of the traditional algorithm in extracting the formant frequency of pathological voice,formant frequency extraction algorithm based on Hilbert-Huang transform and energy separation algorithm is proposed;Aiming at the problem of bandwidth variation of pathological voice signals,the PIF-LPC algorithm is used for bandwidth correction.(3)In terms of speech synthesis and evaluation indicators,according to the repaired pitch and formant,the Klatt speech synthesizer based on the source-filter theory is used for speech synthesis,and objective indicators for evaluating speech quality are studied to evaluate the repair algorithm and system.By comparing the parameters of the speech signal before and after repair,this thesis makes an objective evaluation of the various parts and the overall performance of the repair system.The experimental results show that the excitation parameter repair part can extract accurate and stable pitch information;The vocal tract parameter repair part can obtain the formant envelope with accurate frequency and normal bandwidth;The pathological voice repair system performs well on the four objective indicators of jitter(local),jitter(ppq5),shimmer(local),and shimmer(apq11).The pathological voice repair algorithm based on the nonlinear model proposed in this thesis can greatly improve the quality and intelligibility of pathological voice.