基于多模态音视频融合的质量评价算法  

Quality evaluation algorithm based on multi-modal audio and video fusion

作　　者：袁同庆[1,2] 席鹏 YUAN Tong-qing;XI Peng(Institute of Intelligent Education,Anhui Normal University,Wuhu 241000,China;College of Educational Sciences,Anhui Normal University,Wuhu 241000,China;Suzhou Research Institute,University of Science and Technology of China,Suzhou 215000,China)

机构地区：[1]安徽师范大学智能教育研究院,安徽芜湖241000 [2]安徽师范大学教育科学学院,安徽芜湖241000 [3]中国科学技术大学苏州研究院,江苏苏州215000

出　　处：《沈阳工业大学学报》2022年第3期331-335,共5页Journal of Shenyang University of Technology

基　　金：全国教育科学“十三五”规划教育部重点课题(DCA160264).

摘　　要：针对传统客观评价方法中仅采用调查问卷的数据进质量评价和分类,存在数据不足和相对片面的问题,提出了一种基于多模态音视频融合的客观质量评价算法.该算法充分考虑了客观质量评价过程中产生的视频、语音和文本数据,并分别提取了各类数据中与评价结果相关的特征,通过对多模态数据特征进行加权融合和分类后,得到客观评价结果.以教学质量评价为例,采用自行搜集整理的客观评价数据构建的质量评价数据集进行评估分析.结果表明,与现有方法和基于调查问卷的传统方法相比,所提方法能够明显提升客观质量评价的精度.Aiming at the problems of insufficient data and relative one-sidedness in traditional objective evaluation methods,which only use questionnaire data for quality evaluation and classification,an objective quality evaluation algorithm based on multi-modal audio and video fusion was proposed.Video,voice and text data generated in the objective quality evaluation process were fully considered by this algorithm,and the features related to the evaluation results were extracted from various types of data.After the weighted fusion and classification of the multi-modal data features,the objective evaluation results were obtained.Taking teaching quality evaluation as an example,the quality evaluation data set composed of self-collected objective evaluation data was used for evaluation and analysis.The results show that the as-proposed method can significantly improve the accuracy of objective quality evaluation,compared with the currently existing methods and the traditional ones based on questionnaires.

关 键 词：客观评价 质量评价 多模态特征 语音 视频 文本分类 特征提取 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]