Binaural Asymmetric Directional Microphone Algorithm with Automatic Switching Mode for Better Noise Reduction

Abstract

Noise reduction algorithm on digital hearing aid has been used to improve speech quality and intelligibility at noisy signal. Noise reduction algorithm can be classified by the number of microphone; single channel- or multichannel- noise reduction. At the previous studies about the improvement of speech quality and intelligibility in noisy situation, single channel algorithm is only improved the performance of speech quality by normal and impaired hearing person. However, multichannel algorithm is improved the both performances. The algorithm that has two or more microphones is called a beamforming or spatial filtering. In order to change a directionality of input signals, the algorithm adjusts a phase difference corresponding to a distance of both microphones by additional delay according to the direction of input signal and controls relative amplitude of the input signal at each microphone by a constrained weighting function according to the direction of input signal. As a result, the directionality changes according to the direction of input signal. When the beamforming is used hearing aid, it has several weaknesses. First, the number of microphone is limited by size of hearing aid. That is why the directionality is constrained to the front or rear. Second, manual or automatic switching between non-directionality and directionality is uncomfortable for the users. So many users often do not use the algorithm. Third, the beamforming, that distort the input signal according to the direction, disrupts the brain functions when the direction of the mismatch between the desired signal and input signal is occurred. Human can listen to the speech or desired signal in noisy situation without any other device. It is because the brain functions such as Auditory Scene Analysis and auditory attention. Its phenomenon is demonstrated by the cocktail party effect. In other word, center auditory system of human’s brain play a sound source separator role in noisy situation. Most hearing impaired persons still have the brain functions. Even if an interest signal is front, beamforming has benefit. But if that is the others direction, the algorithm distort the input signal that want to hearing. That is because the brain functions can be disrupted through the beamforming. In order to overcome the above mentioned weaknesses, bilateral asymmetric directional microphone fitting is presented. The method has different directionality at each ear in order to maintain the brain functions. Left ear has directional mode to enhance the front signal, such as beamforming, and Right ear has omnidirectional mode to receive equal signals regardless of direction. Because of the input signal of right ear do not distort according to input direction, the brain functions connected right ear are better performance than that of left. Thereby, left ear has a benefit of directional mode. And right ear has a benefit omnidirectional mode that maintains the brain functions. Also the method removes the necessity for manual or automatic switching between non-directionality and directionality. That is why the method has both directional and omnidirectional mode. As above mentioned, the asymmetric directional microphone fitting has several benefits. But, a dominant noise, when locate in the side of directional mode, may be reduced. However, if a dominant noise locates in the side of omnidirectional mode, its noise may not be reduced. In other words, the performance of the method is expected to be dependent on the according to the direction of the input noise. That is because the directionality modes are not switched according to the direction of input noise. In my study, novel binaural asymmetric directional microphone algorithm was developed and evaluated. The proposed algorithm operates automatically directionality modes switching between left directional and right omnidirectional mode according to the input signal-to-noise ratio. When a dominant noise locates in the side of omnidirectional mode, the proposed algorithm detects the direction of the dominant noise and switches the directionality from omnidirectional to directional mode. MATLAB and Simulink are used to implement the algorithms. For the evaluation, the performance of detection of the dominant noise on the proposed algorithm was measured on 3 kind of scenario (non-reverberant, reverberant, realistic reverberant) by personal computer simulation. And the comparison of performance between the proposed algorithm and conventional asymmetric directional fitting was executed on personal computer simulation and KEMAR test. Finally, the proposed algorithm was tested by normal hearing person. The output signal-to-noise ratio, Perceptual Evaluation of Speech Quality (PESQ) and Hearing Aid Speech Quality (HASQI) are used as objective measurement. As subjective measurement, 10 normal hearing persons are tested by Comparison Mean Opinion Score (CMOS). The result of detection of dominant noise is higher accurate at all scenario (non-reverberant > realistic reverberant > reverberant). As result of objective measurement on personal computer simulation and KEMAR test, when a dominant noise locates in the side of directional mode, the objective measurements of the proposed algorithm is similar to conventional algorithm. But the objective measurements are higher than that of the conventional algorithm, when a dominant noise locates in the side of omnidirectional mode. Especially, objective measurement relative to intelligibility is high at KEMAR test than at personal computer simulation. At the result of CMOS on normal hearing person, the subjective quality of normal hearing person is slightly better on the proposed algorithm. In my study, novel binaural asymmetric directional microphone algorithm that automatically switches the directionality modes between both ears and adjusts the directionality on the one side ear according to the direction of input signal was presented. Through the evaluation on person computer simulation, KEMAR test, and normal hearing person, the proposed algorithm is better performance of objective and subjective measurement than conventional algorithm. It is more influenced by the automatically switching the directionality modes between left and right ear than by the adjusting the directionality on one side ear. Speech intelligibility and quality of normal hearing person is even high on noisy situation. That may be why the result of CMOS is slightly better on normal hearing person. In my study, novel binaural asymmetric directional microphone algorithm was developed and evaluated. I expect that the algorithm is applied to the hearing aid through additional clinical trials for hearing impaired person is tested.