Research On Near-Field Microphone Array Sound Localization And Enhancement Based On Spatial Domain

With the development of science and technology, the microphone array signal processing has gradually become a major research focus. Based on time domain and frequency domain, space domain is added. In the space domain, microphone array not only has the characteristics of tracking and positioning to acoustic sources from different directions, but also has the function of the spatial filter. Both of the characteristics provide a new way of thinking for the speech signal processing. In this paper, we focus on near-field acoustic source localization and enhancement following the new idea. The main content of the article can be summarized as following:Firstly, the characteristics of speech signal and acoustic wave propagation are analyzed. Combined with the traditional far-field signal model of uniform linear array (ULA), the near-field spherical wave signal mode of uniform linear array (ULA), uniform circular array (UCA) and arbitrary topology are studied.Secondly, in order to address about the pitch angle fuzzy caused by UCA, a new microphone array model is established. In this new model, the wideband focusing algorithm and the near-field3D-MUSIC algorithm are combined. Three-dimensional sound sources are located by this algorithm. According to the computer simulation, the three-dimensional multiple signal classification (3D-MUSIC) algorithm based on wideband focusing not only overcomes the shortcoming of traditional space smoothing method only for ULA, but also has better practically and effectively than the near-field3D-MUSIC algorithm. However, this method has extraordinary complexity. Therefore, the two-step estimation method is used to solve this problem. At last, the computer simulation proves that this method greatly reduces the computational complexity on the basis of ensuring positioning accuracy.Thirdly, when the above acoustic source localization has estimation errors, several near-field beamforming algorithms are analyzed. According to the theoretical analysis and computer simulation, the near-field robust capon beamforming (RCB) algorithm has robustness against both steering vector mismatches and covariance matrix mismatches. Due to low robustness of the fixed beamforming in GSC algorithm, sound source localization estimation error can seriously affect to obtain the sound source of expected direction, the fixed beamforming algorithms of GSC is substituted by the RCB algorithm with robustness. Finally, the desire direction sound source is enhanced, and then the computer simulation proves that the GSC algorithm based on RCB has better speech enhancement effect than other algorithms.