ADSP-BF531在嵌入式语音识别系统中的应用  

Application of ADSP-BF531 in embedded speech recognition system

作　　者：王维强[1] 

机构地区：[1]中国空空导弹研究院,河南洛阳471009

出　　处：《电子设计工程》2012年第12期186-189,共4页Electronic Design Engineering

摘　　要：设计了一个嵌入式语音识别系统,该系统硬件平台以ADSP-BF531为核心,采用离散隐马尔可夫模型(DHMM)检测和识别算法完成了对非特定人的孤立词语音识别。试验结果表明,该系统对非特定人短词汇的综合识别率在90%以上。该系统具有小型、高速、可靠以及扩展性好等特点;可应用于许多特定场合,有很好的市场前景。文中讲述了该系统CODEC、片外RAM、ROM以及CPLD等与DSP的接口设计,语音识别运用的矢量量化、Mel倒谱参数、Viterbi等有关算法及其实际应用效果。Designed an embedded speech recognition system. The platform of this system was centered ADSP-BF531. AND it was realization of speaker-independent＇s isolated word recognition through DHMM. The experiment confirms that its speech recognition accuracy reaches 90 percent for speaker-independent and short vocabulary. This system has advantages of small volume, high speed and high expansibility. It is very convenient for some special situations.It has much market potential.This paper introduced the interface of CODEC,RAM,ROM,CPLD and DSP,then recommended the related soft algorithm of VQ, MFCC, Viterbi, ere and it＇ s application effect.

关 键 词：ADSP—BF531 语音识别 离散隐马尔可夫模型 非特定人 孤立词 

分 类 号：TP274.2[自动化与计算机技术—检测技术与自动化装置]