Research Of The Speech Enhancement Algorithm In Near-field Environment Based On Microphone Array

The rapid development of science and technology puts high demands on speech signal processing technology.Whether signal transmission or machine speech recognition,it is hoped that the voice signal does not contain other noise to improve transmission efficiency and system performance.The traditional single-channel noise reduction technology can't suppress the noise because it can't collect the spatial information of the signal,especially when there are speech noise,room reverberation and other scattering noise.In order to improve the noise removal capability,a multi-channel noise reduction technology based on a microphone array is proposed.When the microphone array is located in the near or far field of the sound field,the voice signal model is different.Therefore,using the far field speech enhancement algorithm in the near field of the sound source will inevitably cause its performance to drop rapidly.Therefore,a speech enhancement algorithm capable of having high performance in the near field environment will be indispensable.In this paper,the near-field and far-field models of the speech signal are firstly analyzed in detail,and the reason why the near-field and far-field signal of the sound field need to be treated differently in the speech enhancement algorithm based on the microphone array is analyzed.Based on above,the array matrix and its advantages and disadvantages of different microphone array topologies in the near or far field of the sound field are further analyzed.Secondly,the design principles of the near-field beamforming algorithm and the conditions under which the optimal weight vector should be satisfied are discussed.Then the next part analyze the angular resolution performance and distance resolution performance of several common criteria(MSNR criterion,MMSE criterion,LCMV criterion)of the beamforming algorithm,and design the experiment to verify their noise suppression performance.Then,according to the limitations of the beamforming algorithm,the principle of the ST-GSC speech enhancement algorithm is further analyzed.Aiming at the shortcoming that the adaptive noise canceler in ST-GSC algorithm can't cancel the noise well,an improved algorithm is proposed.The improvement is specifically to increase the constraint conditions when solving the adaptive noise canceler coefficients,so that it becomes a convex optimization problem.The improved algorithm was verified in different noise environments,specifically was in different rooms which speech noise is located near or far field.The experimental results show that the improved algorithm improves the noise suppression performance.Finally,combining the improved ST-GSC algorithm with the phase spectrum compensation algorithm to further improve the noise suppression performance.A noise power estimation method based on the presence probability of speech is proposed to improve the phase spectrum compensation algorithm,and the adaptive ability of the blocking matrix is adjusted to enhance the robustness of the algorithm.