基于内容的语音课件关键词检索系统:设计与实现  

CONTEXT-BASED KEYWORDS RETRIEVAL FOR AUDIO SPEECH COURSEWARE:SYSTEM DESIGN AND IMPLEMENTATION

作　　者：王霅煜[1] 涂惠燕[1] 

机构地区：[1]上海交通大学计算机科学与工程系,上海200240

出　　处：《计算机应用与软件》2011年第4期120-123,139,共5页Computer Applications and Software

基　　金：国家十一五支撑项目(2007BAH09B05)

摘　　要：为了在远程教育环境中满足多媒体课件的关键词检索需求,描述了一种基于矢量量化(VQ)及连续语音识别(CSR)的关键词检索系统的设计与实现。该系统首先利用矢量量化算法对声学特征空间进行聚类并生成码本;接着利用该码本对语音文件逐帧进行处理并保存若干与该帧特征最相似的码表向量所对应的码值做成特征矩阵;然后利用改进的快速符号查找算法从特征矩阵中找出若干候选结果段;使用经简化的连续语音识别算法对候选段进行验证筛选,从而得到最终的结果。在此基础上利用一些测试数据给出其性能表现并做出分析。To satisfy the needs of keywords retrieval for multimedia courseware in distance education environment,the design and implementation of a keywords retrieval system with vector quantization（VQ） and continuous speech recognition（CSR） as its basis was depicted in the article.In this system,vector quantization algorithm is first used to cluster the feature vectors in the acoustic feature space and generate a code book.Then this code book is used to process original speech files frame by frame and to save the code values of several feature vectors which correspond closest to the original speech signal to form a feature matrix.After that a modified fast symbol search algorithm is used to get a group of candidate sections from the feature matrix.At last,a simplified continuous speech recognition method is applied to verify and sift the candidate sections for the final results.In the last part of this article,we displays its performance with analyses based on the system with some testing data.

关 键 词：关键词检索 矢量量化 符号查找 连续语音识别 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]