Research On Speech Enhancement Based On Small Microphone Array

Speech is the main way of human information communication. With the development of social science and technology, people research the speech gradually further and the information communication is also gradually introduced to various kinds of electronic equipment. The speech signal, in the process of communication, is inevitably affected by outside noise and some other noises, such as the noise from the equipment. Due to the interference of noise will affect the quality of people’s hearing, the method of speech enhancement is adopted to reduce the noise interference. The purpose of speech enhancement is improving the quality and the intelligibility of speech by the means of removing noise components from contaminated voice signal as far as possible to retain target speech signals.Compared with the limitations of the distance between sound source and microphone and the comparatively immovable position of single microphone speech enhancement, the microphone array speech enhancement, can not only overcome the above limitations well, but in the complex acoustic environment, can get more time domain and spatial information of speech signal to suppress noise and get high quality target voice. But the traditional microphone array structure is limited by cost and space factors, it’s not applied to the miniaturization very well. Based on this background, this thesis focuses on the small based on microphone array speech enhancement algorithm, the main work is as follows:Firstly, this paper expounds the fixed beamforming, generalized sidelobe canceller, the principle of correlation filter on the small microphone array model and de-noising performance analysis. And in a given noise environment, it can prove its de-noising performance through the experimental simulation. Compared with the other two algorithms, not only the coherent filter can show very good noise suppression characteristics under the condition of multiple noise sources, but also can eliminate noise well and improve the speech intelligibility in different noise environment.Secondly, in view of inaccurate problem of the coherent filter of pure speech signal power spectrum estimation, this article quotes Jeannes’ improved correlation filter algorithm. This paper transforms the problem of pure speech signal power spectrum estimation to the noise power spectrum estimation, and then combines with Martin’s noise spectrum estimation algorithm that is based on minimum statistics to form a new algorithm, which enhance the de-noising ability of the speech enhancement system.Finally, minimum statistical noise spectrum estimation algorithm causes the spectrum estimation errors because of tracking delay, as a result, it affects the combination algorithm of de-noising performance. For this problem, this paper proposes a minimum statistical noise spectrum estimation algorithm based on the weighted average spectrum. Instead of tracking the noise spectrum without time window, the algorithm, estimates whether the noise spectrum needs to be updated by introducing the threshold, introduces a smooth passage parameters to update the noise spectrum smooth at the same time to make the noise spectrum estimation more accurate.Combined the improved noise spectrum estimation algorithm above-mentioned with the correlation filter algorithm, and applied to the topology of the small microphone array, the paper proposes a spectrum based on weighted average of the noise spectrum estimation and coherent filter combining the small microphone array speech enhancement algorithm. The experimental simulation results show that compared with contrast algorithm, the proposed algorithm can improve the quality of the speech intelligibility and voice well.