论文部分内容阅读
BJD—I汉语理解系统作为一铁路客运服务系统,由语声处理、语言处理、语声应答这三大部分组成。本文论述BJD—I汉语理解系统的语声处理。在介绍语声处理部分基本构成及性能的基础上,提出了一种用于准连续语音音节分割的最大共振峰检测算法MFDS和一种新的快速动态规划匹配算法FSDP.并由实验证明,MFDS算法使准连续语音音节分割正确率达到98%以上。而FSDP同Sako提出的最佳DP算法相比,不仅具有相同的识别精度,而且,模式匹配速度快一倍以上。
As a railway passenger service system, BJD-I Chinese Comprehension System consists of three parts: voice processing, language processing and voice response. This article discusses the phonetic processing of the BJD-I Chinese comprehension system. Based on the introduction of the basic structure and performance of speech processing, a maximum formant detection algorithm MFDS for quasi-continuous speech syllable segmentation and a new fast dynamic programming matching algorithm FSDP are proposed. It is proved by experiments that MFDS The algorithm makes quasi-continuous speech syllable segmentation accuracy of more than 98%. However, compared with the best DP algorithm proposed by Sako, FSDP not only has the same recognition accuracy, but also more than double the pattern matching speed.