中文摘要：

提出了一种基于MFCC系数分析和仿生模式识别的语音识别方法，该方法对训练样本MFCC相同分量在各类语音间距离进行了分析，并通过与传统选取方法的比较实验，说明在小词汇量的语音识别中，选取合适的MFCC系数，不仅能减小计算量，正确识别率也会得到一定程度的提高。运用仿生模式识别理论中同类样本连续的观点，通过在特征空间中对训练样本进行有效的覆盖，大大提高了识别结果。

英文摘要：

A speech recognition method is proposed based on MFCC analysis and Biomimetic Pattern Recognition（BPR）.It analyzes the distances among different samples kinds about same MFCC component and selects right coefficients,the comparative experiments are conducted with conditional method.The results show that, in the condition of small vocabulary, the calculated efficiency and recognition rates are improved.Based on BPR continuous theory, it finds the optimal covering of sample features in the same class,the recognition result is improved greatly.