论文部分内容阅读
针对人机交互过程中语音识别引起的发音变异以及用户表达关键信息不完整情况,提出一种模糊匹配方法.该方法分两步,第一步,通过条件随机场进行序列标注,定位查询语句中的关键语义概念,并得到其初步类别;第二步,利用几种相似度计算方法,寻找与领域词典中发音相似度最大的字符串对错误的语义概念进行替换,并标注出具体类别.另外针对最优模糊匹配结果不一定满足用户需要,进行了多个候选的实验.实验结果证明:无论使用哪种相似度计算方法,基于拼音的模糊匹配方法比基于字的模糊匹配方法在语音识别的文本上都具有更好的性能,而且在多候选的结果上也仍旧适用,说明该方法对于提高口语理解系统的鲁棒性上是有效的.
Aiming at the variation of pronunciation caused by speech recognition in human-computer interaction and the incomplete information of key information expressed by users, a fuzzy matching method is proposed in this paper. The method consists of two steps. In the first step, The second step is to search for the string with the most similar pronunciations in domain dictionaries and replace the wrong semantic concept with the specific category by using several similarity calculation methods. Experimental results show that, regardless of which similarity calculation method is used, the Pinyin-based fuzzy matching method is more effective than the word-based fuzzy matching method in speech recognition Text has better performance, and still applies to the results of multiple candidates, indicating that this method is effective in improving the robustness of the spoken language comprehension system.