语音识别技术在多媒体信息检索中的应用仿真  

Application of Speech Recognition Technology in Multimedia Information Retrieval

作　　者：段维军 DUAN Wei-jun(School of Foreign Studies,Jinan University,Guangzhou Guangdong 510632,China)

机构地区：[1]暨南大学外国语学院

出　　处：《计算机仿真》2019年第7期331-334,384,共5页Computer Simulation

基　　金：模特智能机器人控制系统的设计(MHKJ20151080322-2);智能机器人书面交际处理系统的设计(FAIT20151001-1)

摘　　要：针对当前信息检索方法中存在用户满意度低、检索占用存储空间大和检索效率低的问题,提出基于语音识别技术的多媒体信息检索方法。确定原始多媒体信息数据中条件属性集与结论属性集,并进行简单处理,得到一个决策体系。根据上述体系中的两个属性集合实现多媒体信息数据粗糙分类,对粗糙分类结果进行约简,得到信息属性重要性量度。设置给定目标规则属性针对决策重要度量度,如果信息数据属性重要性量度小于等于目标规则属性针对决策重要度量度,那么将该信息数据属性值清洗掉。据此输入关键词语音,并对输入的语音特征进行提取和划分,通过最大似然判决获取声音种类。假设窗内语音种类特征为服从单高斯概率分布状态,则利用某种测度对与窗内语音相似度最高的信息之间距离进行计算,计算期间引入用户检索历史满意度指数,实现信息检索。实验结果表明,上述方法平均检索时长为15.8μs,用户满意度最高可达99%,存储空间占用较少。At present, the information retrieval method has low user satisfaction, large retrieval storage space and low retrieval efficiency. Therefore, a method to retrieve multimedia information based on speech recognition technology was presented. First of all, the condition attribute set and the conclusion attribute set in the original multimedia information data were determined and processed simply to obtain a decision system. According to two attribute sets in above system, the rough classification of multimedia information data was achieved and the rough classification results were reduced to obtain the importance measure of information attribute. Then, the targeting rule attribute for decision importance measure was given. If the importance measure of information data attribute was less than or equal to the targeting rule attribute for decision importance measure, the information data attribute value was cleaned out. On this basis, the keyword speech was inputted, and then the inputted speech features were extracted and divided. Moreover, the type of sound was obtained by maximum likelihood decision. If the feature of voice type within the window was subject to the single Gaussian probability distribution, some kinds of measures were used to calculate the distance between information with the maximum similarity of voice in the window. Finally, the user search history satisfaction index was introduced during the calculation. Thus, the information retrieval was achieved. Simulation results show that the average retrieval time of above method is 15.8μs and the user satisfaction is up to 99%, which needs less storage space.

关 键 词：语音识别技术 多媒体信息 检索 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]