Arbitrary Microphone Array Calibration And Speech Enhancement

Microphone array can get spatial information of the signal and achieve spatial filtering, which makes it widely applicated in all kinds of speech processing system. While in fact, the existence of the noise and interference reduces the quality of the speech signal microphone array get, and affected the clarity and intelligibility of speech signals seriously. Therefore, it is important and meaningful to study speech signal enhancement. Acoording to present study, the simulation results of most speech enhancement algorithm prove the validity of them. But in practical application, their results are not ideal. It is because that most of them are deeply dependent on the array, especially the subspace and beamforming algorithm, the error of the array will directly affect the performance of enhancement. To ensure the validity of enhancement algorithm in practical application, we also should study the array calibration.According to the project requirements, the thesis makes speech signal which microphone array get as the main object of study. The algorithm about calibration of array element position and the adaptive beamforming speech enhancement are lucubrated and discussed. The main contents and innovative points of this article as follows:Firstly, the thesis introduces the present research situation and development trend of microphone array calibration and speech enhancement, the process and characteristic of speech signal, the classification of noise, and discusses speech quality evaluation criteria.Secondly, the microphone array signal processing model is studied, and the micro-phone array of arbitrary structure model is established based on the traditional array model. Then the article analyzes the common forms of microphone array error, and it proposes a new new near field broadband array error model, and analyzes the errors’ effect on beamforming algorithm by simulation. Also variety of classic array calibration me-thods are studied, and the self-calibration is focused on. By analyze the anvantages and disadvantages of those methods, the thesis study a new self-calibration together with MUSIC algorithm and traditional self-calibration to calibrate the position of microphone, improved the calibration source azimuth inaccuracy’s impact on the algorithm, the simulation proved the effectiveness of it.Finally, as the main emphasis, the basic algorithm about speech enhancement are studied in detail with their advantages and disadvantages. The thesis mainly studied the adaptive beamforming algorithm, time delay estimation method and the performance of different time delay estimation. Analysing the Generalized Sidelobe Canceller(GSC) beamforming algorithm and its performance on eliminate noises. Then based on the theory before, the thesis proposes a new GSC algorithm combined band decompose and rear filter algorithm. The algorithm is suitable for the coherent and incoherent noise field at the same time, extends the scope of application, improves the Signal-to-Noise Ratio(SNR), and reduces the amount of calculation. Also analyzing a blind beamforming algorithm which based on higher order accumulation. Compared with the traditional beamforming, the algorithm has a higher output SNR under the condition of array error.