An Improved Dual-Microphone Speech Enhancement Algorithm Based On The Coherence Function

As an important part of speech signal processing, speech enhancement technology is widely applied in the field of speech processing. There are many different kinds of noises in the practical application environment, and they destroy the intelligibility and quality of the speech. Single microphone speech enhancement technology has obtained many research results, and it is effective for dealing with the additive noises in general environments. But dual-microphone speech enhancement technology has developed rapidly and has become a mainstream noise reduction algorithm gradually as a result of the bad performance of single microphone in the respect of processing disgusting noise.By making full use of the speech spatial information, the dual-microphone speech enhancement algorithm realizes the spatial location of speech signals in order to receive the targeted speech signals and filter the noise. The common dual-microphone speech enhancement algorithms are based on the microphone array, coherence function, phase difference and power level difference. This thesis mainly discusses a kind of speech enhancement algorithm based on coherence function applied to hearing aid, which can effectively reduce the coherent noise in speech. And the improved algorithm makes it more suitable for hand-held terminals application environment.First of all, in order for the algorithm to be used in mobile phones, a series of modifications have been made to the original coherence based algorithm. By the analysis of the original algorithm, the improved dual-microphone speech enhancement algorithm adjusts the direction of the received targeted speech, making it easily to receive speech in the direction of perpendicular to the plane of double microphone, and filter out the interferences signals in other directions. After combining the minimum tracking algorithm to the original spectral subtraction filter, this thesis has gotten a new algorithm which has a better filtering effects on the non-stationary noise. Also an estimation signal-to-noise ratio based wiener filtering is obtained to further filtering the noise in non-speech segment. At last, the analysis results of the fixed smoothing factor tells us that the fixed smoothing factor exists some defects, so this thesis replace it with an optimizes variable smoothing factor that could dynamically adjust the estimated value to approach the true signal value preferably.In order to compared performances of the spectral subtraction algorithm based single-microphone speech enhancement and both original and improved coherence based algorithm for dual-microphone speech enhancement algorithms, some tests have been conducted by simulations of a noisy environment of a mall office. Tests results have shown that the performance of the improved dual-microphone speech enhancement algorithm is much better than that of the original, and that the improved dual-microphone speech enhancement algorithm has ability to detect the speech in the specific direction. As a result,the improved dual-microphone speech enhancement algorithm can be applied to the environment in which the mobile phones are used.