基于FPGA的关键词识别系统实现  

Research on keyword recognition system based on FPGA

作　　者：阙大顺[1] 田犇[1] 赵永安 

机构地区：[1]武汉理工大学信息工程学院,武汉430070 [2]深圳华为海思半导体有限公司,广东深圳518129

出　　处：《计算机工程与应用》2013年第8期217-221,共5页Computer Engineering and Applications

基　　金：湖北省科技攻关计划(No.2007AA101C52)

摘　　要：随着微电子技术的高速发展,基于片上系统SOC的关键词识别系统的研究已成为当前语音处理领域的研究热点和难点。运用Xilinx公司ViterxⅡ Pro开发板作为硬件平台,结合ISE10.1集成开发环境,完成了语音帧输出、MFCC、VQ和HMM等子模块的设计;提出了一种语音帧压缩模块架构,有效实现了语音帧信息到VQ标号序列的压缩,实现了由语音帧压缩模块和HMM模块构建的FPGA关键词识别系统。仿真实验结果表明,该系统具有较高的识别率和实时性,为关键词识别系统的FPGA硬件电路的实现研究提供了实例。With the rapid development of microelectronics, the research of keyword recognition system based on SOC （System on a Chip） has become a hot and difficult topic of speech processing field. This paper completes the module design of speech frame output, Mel Frequency Cepstrum Coefficient（MFCC）, Vector Quantization（VQ） and Hidden Markov Model（HMM）, using ViterxlI Pro board by Xilinx Inc combined with ISE10. 1 integrated development environment. A new structure of speech frame compression module is proposed, which effectively achieves the compression from the speech frame information to VQ label sequence. The Field Programmable Gate Array （FPGA） keyword recognition system is also constructed based on the speech frame compression module and HMM module. The experimental results show that the system has a high recognition rate and good real-time performance. The paper provides an example for studying keyword recognition system based on FPGA.

关 键 词：关键词识别 片上系统(soc) 现场可编程门阵列(FPGA) 隐马尔可夫模型(HMM) 矢量量化(VQ) 

分 类 号：TN912.34[电子电信—通信与信息系统]