Research On Key Algorithms Of Acoustic Signal Processing In Hearing Aids

The number of elderly patients suffered from hearing loss is increasing with the advent of the aging society.The elderly are mentally plagued by hearing loss which is second to digestive system disease.Long-term suffering from age-related hearing loss has a great influence on their physical as well as the mental health,and reduces their quality of life.For those with mild,moderate,even severe hearing loss,wearing hearing aids is the most effective sulution for hearing intervention and rehabilitation.For almost a century,the technology of hearing aids has achieved great progress along with users’ satisfaction from 59.6%in 1980 to 74.0%in 2008.However,the performance is hardly to improve due to the howling,signal distortion and noise.Some patients have fitted but not weared their hearing aids.The main reason is that they benefit insufficiently from them.Therefore,further studies on the technologies of digital hearing aids for the elderly with hearing loss are urgently required.This thesis has investigated some key technologies of the Chinese digital hearing aids based on the loudness compensation,adaptive acoustic feedback cancellation,nonlinear frequency compression and speech intelligibility enhancement.The main content of this thesis is given as follows:1.An adaptive loudness compensation method based on the auditory characteristics is proposed to alleviate the confliction between audibility and distortion in conventional loudness compensation method.Firstly,the signal is decomposed into 16 band-pass signals by gammatone filter banks to mimic the human auditory characteristics.Secondly,compensation method is determined by the hearing range of the hearing impaired and the sound pressures of the channel signals.During the compensation,adaptive wide dynamic range compression(WDRC)is employed if nonlinear compensation method is adopted,in which compression ratio is adaptively adjusted to approximate to 1 according to the residual hearing range of the patient and the sound pressuer of the output signal.The goal of adjusting the ratio is to reduce signal distortion at the premise of audibility.At the same time,adaptive WDRC method can increase the output signal to noise ratio(SNR).Experimental results demonstrate that the output SNR of the proposed method in babble noise is improved compared to that of WDRC compensation method.The speech intelligibility and quality are improved compared to the linear and WDRC compensation method.2.A variable step size algorithm for acoustic feedback cancellation in hearing aids is proposed to alleviate the confliction between the fast convergence rate and the low misalignment in adaptive filter algorithms.In the proposed approach,the normalized distance between the short-term average and the long-term average of the filter coefficients is utilized to classify the filter updating state into the convergent state,the transitional state and the steady state.Different step sizes are employed adaptively in different updating states.A large step size is adopted in the convergent state to ensure fast convergence.In the transitional state,a declining lady-like step size is employed to further lower misalignment.In order to make sure the system can finally converge,a small step size is used in steady state.The variable step size algorithm is applied to Normalized Least Mean Error(NLMS)and Normalized Sub-band Adaptive Filter(NSAF)algorithm,forming Variable Step Size NLMS(VSS-NLMS)algorithm and Variable Step Size NSAF(VSS-NSAF)algorithm,respectively.Experimental results illustrate that the proposed algorithms has fast convergence rate,low steady misalignment and high values of Perceptual Evaluation of Speech Quality(PESQ)of the output speech.3.An adaptive nonlinear frequency compression algorithm is proposed to make the most of the residual audible frequency band and improve the speech intelligibility of the hearing-loss patient.Firstly,signals of which the frequencies are higher than the cutoff frequency are decomposed into critical bands based on BARK scale.Secondly,the global compression ratio is determined according to the cutoff frequency and the maximum audible frequency of the patient.Thirdly,the sub-band compression ratio is adaptively determined by the global compression ratio and the normalized average energy of sub-band components.Thus the compressed frequency range of the sub-band signal is determined.Finally,the high frequency signals are compressed to low frequency bands by mapping method,which makes unaudible high-frequency signal audible.Experimental results of speech intelligibility test demonstrate that when compared to WDRC and nonlinear frequency compression algorithms the proposed algorithm significantly improves the intelligibility of initials and sentences under different signal to noise ratio conditions.4.A new feature called Multi-Resolution Power Normalized Cepstrum Coefficients(MRPNCC)are proposed in this thesis to capture both the local and contextual information of spech,and a speech intelligibility enhancement algorithm based on MRPNCC is proposed to improve the speech intelligiblity of the hearing impaired.Firstly,MRPNCC are used to train the support vector machine(SVM)model.Secondly,time-frequency units are classified by the trained model.The time-frequency units dominated by noise are eliminated,and that dominated by speech are remained and processed by wienner filter.Finally,the filtered time-frequency signals are used to reconstruct the enhanced signal.Experimental results show that MRPNCC feature is superior to the traditional spectrum feature.Experimental results of speech intelligibility tests show that the proposed algorithm significantly improves speech intelligibility of the hearing impaired.