| Microphone Arrays have been widely used in speech applications, such asteleconference, hands-free mobile telephone, hearing aids, etc. It has been proposed asa promising solution that can significantly increase the quality of speechcommunication in the noisy background, and will replace conventional desktop andhead microphones in near future.A novel method of speaker tracking and speech separation based on microphonearray has been introduced in this theme. Its hardware and software systems, based onthe TI DSP TMS320C6711, are discussed. The software implementation of thealgorithm, especially, is detailed.This method is based on the near-field microphone arrays speech propagationmodel. By the means of near-field 2D_MUSIC, the near-field based estimates thelocation of speakers. Utilizing this location information, the method attenuatesunwanted voices by Near-Field MVDR Beamforming technology.This theme focuses on the software implementation. Taking its computationalcomplexity and multiple tasks into account, the real time implementation is the keypoint, which consists mainly of two parts, code optimization and tasks scheduling.The former is composed of the utilization of DSPLib and many kinds of trade-offs inthe implementation. Utilizing DSPLib in the programming, the corresponding codeefficiency can be increased dramatically. The trade-offs between space and time,accuracy and computational complexity, on-chip RAM and off-chip RAM have madethe real time implementation possible. In addition, the tasks scheduling arranged therunning time of different tasks according to their priorities. This helps the wholesystem to work in phases.The testing result in real speech environment has showed that this system is ableto track the speaker and pick up the desired speech signals in real time. The testingdata is also analyzed. Several drawbacks, as well as their causes, have been discussed,and the corresponding plan of improvement has been proposed for higher systemperformance. |
PDF Full Text Request