基于FPGA的语音识别SoC系统解决方案

An Speech Recognition SoC Solution Based on FPGA

基于Actel CoreMP7低成本SoC开发验证平台,完成了以ARM7为核心控制器、马氏距离计算专用电路(MSAC)为协处理器的语音识别SoC的设计与验证。实验结果表明,该SoC系统在Actel ProASIC系列FPGA M7A3P1000上综合实现后,约占用M7A3P1000总资源的39.18%及1KB片内SRAM,完成整个语音识别算法性能比S3C44B0x(ARM7)平台上定点C程序提高了49.78%,充分证明了CoreMP7平台用于SoC设计的可行性和便捷性,以及主处理器配合硬件加速协处理器架构在信号处理领域的优势。

A speech recognition SoC which consists of ARM7-TDMI core and a co-processor called MSAC （Mahalanobis Distance Calculation Module） on an Actel coreMP7 based on low-cost SoC development platform is implemented. The whole SoC system consumes 39.18 percent of the total FPGA resources and IKB-chip SRAM when synthesizing on the Actel ProASIC series FPGA M7A3P1000. The performance of the whole speech recognition flow on the SoC is improved 49.78 percent compared to the fixed point program on S3C44BOx, which proves the possibility and efficiency of CoreMP7 platform, and the advantage of combining main processor and hardware co-processors in the field of signal processing.