Research On Self-Fitting Technology Of Digital Hearing Aids

Hearing is one of the most important way for people to communicate in daily life,but hearing loss will seriously affect the individual's normal life.However,the proportion of people with hearing loss is on the rise.Under the existing level of medical conditions,the use of hearing aids has become the primary means of treatment for hearing loss.The traditional fitting process of hearing aid is relatively cumbersome and highly dependent on hearing expert's experience.In recent years, self-fitting of hearing aids has been lifted gradually.Self-fitting means that patients with hearing loss complete the fitting process by themselves through interaction with related software without hearing expert's participation,which breaks the shackles of traditional fitting process.This paper focuses the self-fitting technology based on interactive genetic algorithm and Gaussian process regression.Based on predecessors' achievement,two complete algorithm processes have been proposed which realizes self-fitting of hearing aids preliminarily.Some research results are concluded as follows:(1) Human auditory system, the cause of hearing loss, hearing tests and treatments are studied in this paper.Two methods of fitting,traditional fitting and self-fitting,are discussed in detail.(2) A self-fitting technology based on interactive genetic algorithm is elaborated.Specific provisions of each genetic operator's operation are made,including chromosome encoding, select operation,genetic crossover and mutation related to the type of hearing loss.Besides,a auxiliary module aimed at accelerating the evolution process is proposed,which takes advantage of the existing data in knowledge base to improve search efficiency.(3) A self-fitting technology based on Gaussian process regression is proposed. In this paper,human auditory perception which can not be measured directly is expressed as a internal response function,and it is modeled by a Gaussian process.After studying the Gaussian process regression theory,a series of theoretical derivations are conducted,and a formula which can be used to predict one's preference over current fitting parameters is finally obtained.In addition,a sequential design based on EI(expected improvement) is done to get the "most meaningful" fitting parameters for one to mark,so as to form an interactive fitting loop.(4) Appropriate interactive modules are designed corresponding to fitting methods mentioned above,including pure tone testing and speech testing. A 16-channel 6-order IIR filter bank is designed to tie in with the above algorithm research.Experimental tests show that this filter bank is good in performance which can be used as the signal processing module of a self-fitting system.Finally,the speech recognition experiment and voice quality test prove that the two methods this paper proposed can be up to self-fitting of hearing aids.