基于DSP的口令式语音CDHMM的实时训练系统  

A CDHMM Real-time Training System of Token-based Speech Based-on DSP

作　　者：吴卅建[1] 李辉[1] 戴蓓倩[1] 

机构地区：[1]中国科学技术大学电子科学与技术系,安徽合肥230026

出　　处：《微电子学与计算机》2005年第2期128-131,共4页Microelectronics & Computer

摘　　要：文章给出了基于DSP的特定人口令式语音CDHMM(连续隐马尔可夫模型)的实时训练系统。在分析了CDHMM训练算法运算特性的基础上,采用TI高性能浮点DSP芯片TMS320C6711成功实现了口令式语音的CDHMM的训练。并对系统的存储量及实时性做了深入研究,为语音识别在嵌入式领域中的应用提供了重要的技术途径。In this paper, one CDHMM real-time training system of token-based speech based-on DSP is proposed. Through analyzing CDHMM training algorithm, this system is implemented successfully in using high performance float-point DSP TMS320C6711 of TI incorporated. At the same time, the system's storage need and real-time performance are discussed. Thus an important approach of applying speech recognition system in embedded devices is proposed.

关 键 词：CDHMM DSP 模型训练 语音识别 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]