Reasearch And Implementation Of Speech Recognition System Based On DSC

With the increasing popularity and development of digital integrated circuit technology, more and more embedded digital controllers come around human’s life. In digital controller system, the sensor data collected is converted to digital signal transmission to the MCU for data processing, and then generate the control signal for the executing agencies. However, as the control algorithm becoming more complex and more control signal precision being proposed, more signal post-processing operations are needed. For these calculate-intensive tasks, a simple MCU is hard to complete these signal processing operations within avalidity time for real-time control system.DSC(Digital Signal Controller), which combines MCU(Micro Control Unit)’s and DSP(Digital Signal Processor)’s characteristics, has the advantages of easy programming, flexible controlling, fast calculating and high accuracy.In this work, a programming model based on DSC is studied, and a speech enhancement and recognition system based on DSC is implented. Speech enhancement technology can suppress interference and noise by two non-directional microphones. Speech recognition can recognize ten digit number,0-9, by the use of open source,pocketsphinx.The main contents and results studied in this paper include the following:(1) The programming model based on the heterogeneous dual-core architecture of DSC has been studied. And the communication, synchronization and task partitioning methods between the dual-core have been summed up. Also, the experiments show the DSC is prior than MCU to deal with complicated control task.(2) The speech enhancement technology based on two microphones has been studied. And the speech enhancement algorithm has been discussed. Also, the problems in the realization of speech enhancement in DSC, such as task partition, file IO, fixing point and optimization, have been analyzed.(3) The open source program pockersphinx has been studied. And the speech recognition algorithm has been discussed. Also, the problems in the realization of speech recogniton in DSC, such as task partition, program tailoring, file IO and optimization, have been analyzed.