Experimental spatial filtering and array calibration for speech enhancement
Michael J. Link, K.M. Buckley · 1994
The use of an array of microphones mounted on a mannequin's head used as a preprocessor for a hearing aid is investigated in this thesis. The advantage of using an array of microphones is the user is offered spatial selectivity not available with currently used hearing aids. The processor investigated is a real-time, adaptive, time domain processor. The adaptive processor is used since it offers the advantage of being near optimum in many environments. The performance of the adaptive processor is compared to that of a single microphone by demonstrating the improvement in the signal to noise ratio and an intelligibility based measure. The processor is a real-time system and the response of the microphone array must be known or estimated at various locations. In the thesis a new calibration technique is described and demonstrated via simulations and experimental evaluation. The calibration technique uses measured estimates of the array response information at known directions of arrival to estimate the array response at unmeasured locations. The advantage of the calibration technique demonstrated in the thesis is the response of a head mounted microphone system is not invariant over the field of view. The calibration technique uses a polynomial to fit the model to the sparsely measured responses over the field of view. The polynomial is then used in conjunction with the modeled response to estimate the response at unmeasure directions of arrival. The estimated response of the microphone array is then used to implement the adaptive algorithm.