Research On Microphone Array Speech Enhancement Method

The voice collected from all kinds of equipment is not pure voice in real life, which is often subjected to various kinds of noise interference. Therefore, the speech enhancement method is needed to recover the original speech from the noisy speech as far as possible. The traditional method of single microphone uses the information of time and frequency domain of the signal, which can also suppress the noise, but it has the disadvantage of poor effect for non-stationary noise suppression capability. Now, the method of microphone array uses the information of space and time-frequency domain, which can well solve to this problem of the non-stationary noise suppression.The method of single microphone speech enhancement and array microphone speech enhancement is deeply studied in this paper, and simulation of the method is carried out. Generalized side-lobe canceller algorithm is microphone array speech enhancement method of widely used, when the input signal energy is larger, adaptive noise cancellation module in the generalized side-lobe cancellation algorithm will produce noise amplification problem. An improved generalized side-lobe cancellation algorithm is studied in this paper, the adaptive noise cancellation module in the improved algorithm use a normalized least mean square filter to overcome noise amplification problem, which has better stability and better convergence performance. Due to the generalized side-lobe cancellation algorithm is not suitable for the elimination of non-coherent noise, therefore there is some residual noise when use generalized side-lobe canceller for speech enhancement. At this time, we can use the single microphone speech enhancement method to enhance the speech that has been enhanced by generalized side-lobe canceller algorithm, which can achieve better results. Therefore, this paper proposes an array beam forming speech enhancement method that combines with single microphone, which can eliminate both coherent and non-coherent noise in noisy speech signal. Simulation test results show that the putout speech has a higher signal to noise ratio according to comparing with the traditional generalized side-lobe cancellation algorithm and the speech quality has been significantly improved.