The Design Of Microphone Array Speech Enhancement Based On Beamforming

With the rapid development of semiconductor technology and continuous enhancement of processors, speech processing technology with Microphone Array has substituted for the technology with single channel. Meanwhile, Microphone Array has also been widely used for pickup in many kinds of speech processing equipment such as smartphone, interphone, digital hearing aids etc.Nowadays, high-end digital hearing aids have clear advantages in the market which have become even more obvious especially in recent years because of its strong manipulated ability. The advantages of digital hearing aids of low noise, energy conservation, small size, adaptive adjustment etc. which imitating hearing aids lack bring hope to those with hearing impairment. The widely used Microphone Array in digital hearing aids makes pocket hearing aids become the mainstream.This project focuses on the Microphone Array Apeech Ehancement algorithms suitable for digital hearing aids based on Beamforming technology. In the project, we concentrate on Convention Beamforming algorithm, LCMV Beamforming algorithm and design an adaptive Beamforming algorithm based on LMS by finishing every part and promoting the speech signal. After designing the Microphone Array speech enhancement algorithm, we fully test it and deeply analyze the advantages as well as disadvantages and find that the signal processed by Convention Beamforming has low speech quality and a limited de-noising effect, however, the high-performance LCMV(Linearly Constrained Minimum Variance) Beamforming algorithm needs too much calculation to apply for digital signal. The experiment results show that LMS-GSC algorithm we design has a much higher signal to noise ratio than conventional one and in the meantime it keeps high speech quality which is more suitable for hearing aids from the perspective of calculation.It's necessary to design a speech detecting algorithms called VAD(Voice Activity Detection) with less calculation and more accurate result because much more efficient VAD can directly result in better conclusion.In order to increase the accuracy of speechdelay compensation the shortest path algorithms are used on the base of Cross-Correlation function to minimize the noise from other directions and maximize the signal from the aim direction. The project also designs the data flow and software with a detailed analysis of the basic operation, discusses the problems in the procedure and optimizes the program.Finally, the improved algorithm is successfully realized on embedded development board based on ARM Cortex-A8 and the algorithm processing effect is good with simulator validation.