Research And Implementation Of Transient Noise Suppression Algorithm For Digital Hearing Aids

Digital Hearing Aid users live in a variety of environments,the same as normal people,full of various noises.The Hearing Aid users' experience depend on the performance of digital hearing aid.The most important performance of the digital hearing aid is the property to suppress noises.There are already many robust algorithms to deal with the stationary noises such as gaussian white noise while a few for transient noises.Typical transient noises,e.g.door knocks,hand claps,metal glasses crashes,electrical impulsive crackles,are short in time,widely spread across the frequency domain,and have an abrupt nature.Transient noises often last about several milliseconds after its appearance and hard to suppress because of its widely spread in frequency domain.Transient noises always have a deadly influence on the performance of digital hearing aid with harming hearing aid users' health twice.We mainly aim at research and implementation of transient noises suppression algorithms for digital hearing aid.In this dissertation,we present two algorithms that suppress different kinds of transient noises.One is Optimally Modified-Log Spectral Amplitude(OM-LSA)which comprises derivation of Optimally Spectral Gain Function and Minima Controlled Recursive Averaging(MCRA)based on priori SNR,posteriori SNR,speech presence uncertainty and noise spectral estimation.After get the Optimally Spectral Gain Function,we use IFFT to get the pure speech.The other is algorithm about detection and suppression of transient noises in acoustic signals using an Exponential Weighted Least Square(EWLS).These two algorithms are for different kinds of transient noises.In order to get a better performance of estimation and suppression for transient noises,we use OM-LSA algorithm to deal with the normal transient noises with common natures.As for the special case of transient noises,we use EWLS algorithm to process impulse noise with temporal smoothing method for more rapid result.We employ the algorithms on both speech and music pieces contaminated by various transient noises in this dissertation.We use Goldwave(Audio Software)to observe and analyse the transient noises both in time and frequency domain after processed.Meanwhile,Spectrograms,Seg SNR,MOS(Mean Opinion Score)are three common criteria to evaluate the performance of relevant algorithms.The result shows that the proposed algorithms display a very good performance for suppressing the signals contaminated by normal transient noises without noise residual and signal distortion.