Study And Realization Of Speech Enhancement Algorithm Based On Dual Mini Microphone Array

Dual mini microphone array is a set of two mini microphone arrays proposed by our research group based on the human ear structure, the most important feature of dual mini microphone array is include short spacing elements and long spacing elements at the same time. The speech enhancement algorithms that developed for dual mini microphone array can be effectively applied to the scenarios such as binaural hearing aids, artificial robot etc. In this paper we had undertaken some work based on dual mini microphone array, those work include:1) Introduced the structural characteristics of dual mini microphone array and derived its array manifold and gave the geometric direction of arrival estimation method(DOA) of dual mini microphone array in free field.2) Combined with the structural characteristics of dual mini microphone array element we proposed coherent filtering generalized sidelobe canceller(CF-GSC) algorithm, it can get better enhancement effect than some traditional algorithm with dual mini microphone array structure. Also a speech enhancement methods with broadband minimum variance distortionless response algorithm based on improved wiener filter was proposed, which can suppress the interface signal of non-target direction effectively. And the modify wiener filter can improved the intelligibility of enhanced voice through reduce the distortion caused by traditional wiener filtering efficiently.3) Combined with the front-end speech enhancement algorithm, discussed how to preserve the binaural clues in dual ear scenarios with dual mini microphone array, that means preserve the interaural time difference(ITD) and then output. Those can effectively promote the hearing experience of hearing loss patient, also ensure the patient can get the perception of ground environment during speech enhancement process, particularly the output of noise clues may help patients avoid external danger.4) Combined with the result of a simply investigation about location and number of microphone on mobile devices, we designed a hearing aids application based on Android mobile platform, used spectral subtraction algorithm to obtain enhanced voice in the device contains only a single microphone, and use the two channel LMS algorithm to get the enhanced speech in the device which contain two or more microphone. Also has developed the frequency compression shift function in this application in consideration of the circumstances of the patients who suffer high-frequency hearing loss. Actual audition test showed the applications can achieve better hearing aid effect under simple noise scenarios.