| In the noisy background, the environmental noise and speeches from other speakers make the speaker of interest cannot be heard explicitly, which makes bad effect to voice communication.Researching of microphone array is a new branch in the field of array signal processing. It brings in new solutions to the above problem by adding a spacial domain to the original time and frequency domains, which makes up the shortages of a single microphone, and can be used in the noisy background to improve the quality of voice communication.A novel method of speaker tracking and speech separation based on microphone array is introduced in this thesis. Its hardware and software systems, based on the TI DSP, are discussed.This method is based on the near-field microphone arrays speech propagation model. First, the average covariance matrix in frequency domain is gained by means of frequency focus technology. Then speaker location can be estimated by near-field wideband 2D-MUSIC algorithm. Utilizing this location information, unwanted voices can be attenuated by Near-Field MVDR Beamforming technology.The hardware and software implementation is discussed. A general hardware platform for array speech processing with TMS320C6713 as the core is implemented. A software system is also designed and optimized, taking account of the multi-tasks and complexity computation. The main difficulties are task schedule, code optimization and microphone calibration. Task schedule arranges the running time for the tasks according to their priority. Code optimization includes two parts. DSPLIB is used to improve the efficiency and several trade-offs are also taken into account. Microphone calibration makes the amplitude response of all microphones to be identical.The testing result shows that this system can track speaker and pick up the desired speech signal in real time. The testing data is analyzed and the plan of future improvement has been discussed as well. |
PDF Full Text Request