【摘要】 本文介绍一个新的阶层单调识别技术。它立足于应用动态规划匹配法时匹配距离的收敛性。在第一阶段用窄窗孔匹配选出候补单词,在第二阶段用宽窗孔匹配,从其中选出正解单词。它不需要为两个阶段分别计算特征参量。与非阶层式相比,它可以节省约78%的识别时间。第一阶段取10%的候补单词时可做到无遗漏,保持了非阶层式的识别精度。在这一点上,它优于用粗分帧平均特征的阶层式单词识别法。更多还原

【Abstract】 In this paper a new technique-a hierarchical method for recognition of iso- lated words is introduced. It is based on the astringency of the DP matching distance. During the initial stage of rude DP-matehing-eandidate words are selected through a narrow matching window. During the second stage these candidates are matched through a wide window, and the proper word is selected as the result of recognition.This method will save 78% the recognition time as compared with the non-hierarchical method. If 10% the candidates is selected, no omission will be found in the first stage, The recognition accuracy of the non-hierarchical method remains in the new method. The experiments and the selection of optimal candidate data are also described in this paper.更多还原