Correction Method Of The Microphone Array

Traditional single microphone systems only process signals in time and frequency domains. Microphone array systems introduce space information of sound sources, i.e. they can additionally process signals in space domain. Thus, they can realize sound source localization, restraint of directional interference, speech separation and so on. They can be widely applied to hearing aids, video conference, speech recognition, multi-media classroom, battlefield interception and so on. Now, speech processing algorithms based on microphone arrays emerge in endlessly. Simulation results have verified effectiveness of these algorithms. In practice, however, performance of these algorithms such as subspace localization algorithms and adaptive beamforming algorithms is dissatisfied.In order to improve the performance of these algorithms in practice, this paper performs preliminary study on this problem. First, this paper illustrates a typical sound source localization algorithm and a typical speech enhancement algorithm of microphone arrays, which are considered as basis of analysis of performance. Then, this paper investigates reasons that make performance of these algorithms degraded in practice, i.e. microphone array errors. According to viewpoints and methods of system, microphone array errors are modeled. On this basis, an improved array model that contains microphone array errors is proposed. Then, this paper investigates the solutions, i.e. calibration approaches of model errors of microphone arrays. According to the status of locations of calibration sound sources, the calibration approaches are divided into active-calibration and self-calibration. Using subspace characteristics, an active-calibration approach is proposed. Then, a robust sound source localization approach on the basis of the active-calibration approach is proposed. Using symmetry, a self-calibration approach is proposed. Finally, comparisons between the approaches and conclusions are given.The contributions of this paper are as follows.(1) An improved model of microphone arrays is proposed. It takes microphone array errors of gain, phase, and element positions into account. It can be applied to planar arrays with arbitrary geometry and can be extended to multi-dimensional microphone arrays. Compared with the traditional model of microphone arrays, it is more approximate to actual situations.(2) An active-calibration approach is proposed. It can calibrate the gain, phase, and position errors of microphone arrays with high accuracy. In order to solve the problem of phase ambiguity in the active-calibration approach, a subspace model error estimation algorithm is proposed. And then, a robust sound sources localization algorithm based on the model error estimation algorithm is proposed. It retains good performance even when model errors exist.(3) A self-calibration approach is proposed. It does not require locations of calibration sound sources. It can calibrate gain and phase errors of microphone arrays using symmetry. The characteristics of the approach are good performance, less number of computations, and facility of using.