摘要： 给出了一个嵌入式小词汇量非特定人语音识别系统的方案，它基于单片微控制器加数字信号处理器（MCU＋DSP）架构，用离散隐马尔可夫模型方法实现语音识别系统。该系统适应性强，可扩展性好，具有一定的实时性和语言无关性。当在算法上针对汉语普通话进行优化后，则能进一步提高识别汉语的准确性和实时性。

关键词: 语音识别；非特定人；隐马尔可夫模型；Mel 倒谱系数；汉语普通话识别

Abstract: This paper introduces a solution of real-time speech recognition in embedded system which is based on MCU + DSP architecture, it realizes the system by hidden markov models (HMM) algorithm. After optimization, the system has a better performance in mandarin recognition.

Key words: Speech recognition; Speaker-independent; Hidden markov models; Mel frequency cepstrum coefficient; Mandarin recognition