第36卷第7期 计算机仿真 2019年7月
文章编号:1006-9348 (2019)07-0331-04
语音识别技术在多媒体信息检索中的应用仿真
段维军
(暨南大学外国语学院,广东广州510632)
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15.8p.s,用户满意度最高
可达99%,存储空间占用较少。
关键词:语音识别技术;多媒体信息;检索
中图分类号:TP391 文献标识码:B
Application of Speech Recognition Technology in
Multimedia Information Retrieval
DUAN Wei-jun
(School of Foreign Studies, Jinan University, Guangzhou Guangdong 510632, China)
ABSTRACT : 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 technolo
gy was presented. First of all, the condition attribute set and the conclusion attribute set in the original multimedia in
formation data were determined and processed simply to obtain a decision system. According to two attribute sets in a-
bove 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 l