Research On Noise Reduction Of Digital Hearing Aids

Hearing is an important part of human communication with others. In recent years, because of the environmental impacts, the number of patients with hearing loss has increased, which has been a serious impediment to the improvement of the national health level and the development of society. For patients with hearing loss, wearing a hearing aid is the best solution. Compared with normal hearing people, because of the damaged hearing organs and the loss of speech recognition, the patients with hearing loss have lower speech comprehension in the noisy environment. The noise reduction technology can improve the signal-to-noise ratio of the speech, and then improve the patients’understanding degree. Therefore, the noise reduction is an important and essential part of digital hearing aids, which is of great significance to improve the performance of digital hearing aids.Supported by the National Natural Science Foundation of China and the Natural Science Foundation of Jiangsu province, the research is mainly focused on the noise reduction methods applied to digital hearing aids. In this paper, the main work is divided into the following parts:(1) This thesis studies the noise power estimation methods based on VAD and the methods based on recursive averaging to update the noise spectrum estimation successively, and introduces the improved spectral entropy method, the MS method and the MCRA method and makes detail analyzations and comparisons about them. The simulation results show that the MCRA method can accurately estimate the noise power spectrum, and so it is used to estimate the noise power spectrum in this paper.(2) In order to minimize the speech distortion, the methods based on the human auditory masking effect are studied. This thesis introduces the human auditory masking effect and the noise reduction method based on the perceptual filter. For general perceptual filters bring the problem that the noise below the masking threshold can become audible after the masking filtering, which is called the maskee-to-audible-noise (MAN) phenomenon, an improved perceptual filter method is studied to avoid this problem. By adding a weighting factor, only the noises which meet the MAN problem are further filtered, thereby eliminate the MAN problem without introducing much distortion. Besides, to get more accurate masking thresholds, the wiener filter is used to estimate the rough spectrum of clean speech rather than the spectral subtraction. Experimental results shows that the proposed method effectively removes the residual noises, reduces the interference of noise on speech, and makes the voice quality better.(3) This thesis studies the noise reduction method based on directional microphones. The research is mainly focused on the first-order adaptive directional microphones and the second-order adaptive directional microphones. For the first-order directional microphone’s directivity index is not high and the second-order directional microphone has the serious white noise gain problem in low frequency band, the paper studies the improved hybrid directional microphone method by combining the advantages of both. At the low frequency band where the frequencies are less than 1000 Hz, the first-order directional microphone is used to avoid large microphone noises, and at the high frequency band where the frequencies are more than 1000Hz, the second-order directional microphone is used to obtain high directivity index. Experimental results, including objective and subjective measurements, show that the hybrid directional microphone method has an obvious improvement.(4) Apply the noise reduction algorithm based on the auditory masking to a hardware platform and use a software running on PC to control the whole system. Experiments based on the hardware platform verifies the validity of algorthms proposed in this paper.