Researches On Hearing Aids' Fitting And Loudness Compensation Methods

Loudness compensation is the fundamental function of hearing aids.How to help hearingimpaired people to hear and understand sounds better by loudness compensation has always been the focus of researches on hearing aids.In recent decades,researchers have been continuously carrying out in-depth researches on many relevant subjects,such as the principles and characteristics of the human auditory system,the physiological mechanism of hearing loss,etc.,and presented many research results,including various audiological models,loudness and voice intelligibility measurement.Although some earlier ones of these results have already been applied to hearing aids' gain prescription and loudness compensation and proved to be beneficial,many later ones have not.Meanwhile,deep learning has achieved great success on many problems in recent years.But due to the lack of data and some other reasons,the application of deep learning in hearing aid fitting has not received much attention.Furthermore,there are many shortcomings in the current hearing aid fitting process.For example,general hearing aids prescription procedures,such as NAL-NL1/2,are widely used in hearing aid fitting,while the prescription obtained by these procedures is just the gains for longterm average speech spectrum(LTASS),rather than which for various sound spectrum.Furthermore,these prescription procedures never take reducing output level as object.Since higher level leads to higher power consumption,it will be of great significance if hearing aids are enabled to provide the same,or even higher,intelligibility at lower output levels.Besides,whether and how deep learning can be applied to hearing aid fitting and are remain to be studied.Under this situation,this paper carries out some researches.The main works of this paper are as follows:(1)In order to provide better loudness compensation effects for various input spectrums,a loudness compensation model that establishes the relationship between gains and input sound spectrum is presented basing on a recently proposed cochlear hearing loss model,by which appropriate gains for various spectrum can be determined.(2)In order to enable the hearing aids to provide the same or even higher intelligibility at lower output sound level for reduce power consumption,the model is optimized for both the intelligibility and level of output sound using genetic algorithm.Based on the optimized model,a complete fitting and loudness compensation procedure is proposed.(3)In order to study the feasibility and methods of the application of deep learning in hearing aid fitting,a gain prescription model based on deep neural network is presented.Its ability to fit the NAL-NL2 prescriptions is verified,and the impact of identity information of users on the performance of the model is studied.