摘要
关键词检测技术是语音识别领域中一个重要的研究方向。对于关键词检测系统来讲,若要求达到高检测率的同时希望虚报率较低,仅仅建立关键词模型是不够的。填料模型的结构和类型对整个系统的性能相对会有更大的影响。对于非特定说话人的连续语音中出现的非关键词语音提出建立一种新的有效的基于词表树结构的填料模型。实验结果表明,与传统的基于音节格和音节聚类的填料模型相比,关键词的检测率有了很大的提高,系统的综合性能较好,具有一定的可行性和实用性。
Keyword spotting is one of important study way in speech recognition.For the keyword spotting sys-tem,if you want to get high detection and low false report,it isn t enough to only construct a keyword model,as the structure and type of fillermodel has an important effect on the overall system performance.A new and efficient fill-ermodel based a lexical tree structure is described,which aims at non-keyword speech that present to continuous speech for non-special speaker.Compared with the trad itional fillermodel based syllable lattice and syllable cluste-ring,experimental results show that the filler model based a lexical tree structure has a great improvement in the probability detection for the keyword,the whole system has a better capability,having the feasibility amd the prac-ticability properly.
出处
《科学技术与工程》
2011年第13期2967-2970,2976,共5页
Science Technology and Engineering
基金
黑龙江省黑龙江科技学院引进高层次人才科研启动基金项目(06-132)资助
关键词
关键词检测
填料模型
词表树
音节格
音节聚类
keyword spotting filler model lexical tree syllable lattice syllable clustering