基于音频特征的收视率调查系统及检索算法研究
发布时间：2019-03-08 13:24
【摘要】：收视率调查仪是继传统的问卷调查收视率后发明的调查收视率的设备,在广电部门普遍应用,具有较高的研究价值。收视率调查结果对于电视台更好的编排节目、广告商适时投放广告、收视群体更好地接收节目信息具有重要的指导意义。目前的收视率调查仪设备的原理多样化,广电部门应用的设备成本较高,并且设备的可靠性有待提高。鉴于这种背景,本文研究了基于音频特征的新型收视率调查设备及其相应的音频检索方法。 本文首先简要介绍了收视率、收视率调查等的相关知识,阐明了收视率调查的主要指标、收视率调查的方法和意义,然后概述了当前主要的收视率调查设备种类、原理及其存在的问题,最后提出了本文研究重点和内容---基于音频特征的收视率调查系统和收视检索算法,并阐述了音频信号处理的方法和音频检索的主要技术。 基于音频特征的收视率调查系统,其整体架构是由用户端的音频数据采集部分、中心端的数据处理部分以及两者之间的通信模块构成,文中论述了这三部分的具体功能及实现时需要解决的问题,并给出了可行的实施方案。 本文重点研究基于音频特征收视率调查系统的有效的特征矢量和音频检索方法,以便降低设备的成本、提高设备的可用性和可靠性。本文设计的音频特征是针对实际电视收看时音频获取中存在的问题而改进的音频特征,采用通带划分技术,并且剔除对高低音响敏感的通带,同时,采用归一化方法避免不同音量对获取的音频特征的影响,从而获得不受电视机音量、型号等影响的、有很好的鲁棒性的音频特征。进而,建立模拟收视调查实验环境,利用MATLAB软件,分析和研究了音频特征的可靠性,以图表方式给出实验结果。 本文设计的音频检索算法分两步进行,首先进行相关性计算,计算待测音频与音频库中相应时间段的音频的相关系数,并采用阈值判定的方法,将相关系数小于阈值的音频剔除；然后进行欧氏距离比对,将相关系数大于阈值的音频继续与音频库中相应时间段的音频进行距离计算,取其距离最小者对应的频道号作为判别结果,同时,采用多包数据判别方法,对用户收看的每一频道取两到三包数据,从而保证判别结果的准确性。最后,本文对所提出的音频检索算法进行了评估,验证其有效性。 作为一种新型收视率调查系统,本文所取数据的特征均是建立在新建系统的基础上,所取数据满足数据量小、有效性高的特点,从而能达到降低用户端设备的复杂度、通讯花费少以及简化中心端数据处理的复杂度等,且通讯模块具有多模式接口,以满足电话传输、无线传输及互联网传输。论文最后对本文所做的工作进行了总结,并指出了今后改进之处,以及下一步研究方向。
[Abstract]:Rating survey instrument is a kind of equipment invented after the traditional questionnaire survey. It is widely used in radio and television department and has high research value. The result of ratings survey is of great significance for TV stations to arrange programs better, advertisers to place advertisements in good time, and television groups to receive program information better. At present, the principle of rating survey equipment is diversified, the cost of equipment applied in radio and television department is high, and the reliability of equipment needs to be improved. In view of this background, this paper studies a new rating survey device based on audio features and its corresponding audio retrieval method. This paper first briefly introduces the relevant knowledge of ratings, ratings survey, etc., clarifies the main indicators of the ratings survey, the method and significance of the ratings survey, and then summarizes the main types of ratings survey equipment at present. At last, this paper puts forward the focus and content of this paper-the ratings survey system based on audio feature and the algorithm of ratings retrieval, and expounds the methods of audio signal processing and the main techniques of audio retrieval. The overall architecture of the ratings survey system based on audio features is composed of the audio data acquisition part at the user end, the data processing part at the central end and the communication module between the two. In this paper, the concrete functions of these three parts and the problems to be solved in their implementation are discussed, and the feasible implementation scheme is given. This paper focuses on the effective feature vector and audio retrieval methods based on the audio feature rating survey system in order to reduce the cost of the equipment and improve the availability and reliability of the equipment. The audio feature designed in this paper is an improved audio feature aiming at the problems existing in the audio acquisition of the actual TV viewing. The pass band division technology is adopted, and the pass bands sensitive to the high and low audio are eliminated. At the same time, The normalization method is used to avoid the influence of different volume on the acquired audio features, so that the audio features that are not affected by the volume and model of the TV set are obtained, and the audio features with good robustness are obtained. The reliability of audio characteristics is analyzed and studied by using MATLAB software, and the experimental results are given in the form of graph. The audio retrieval algorithm designed in this paper is carried out in two steps. First, the correlation calculation is carried out, and the correlation coefficients between the audio to be tested and the corresponding time period in the audio library are calculated, and the threshold decision method is adopted. The audio frequency whose correlation coefficient is less than the threshold value is eliminated; Then the Euclidean distance comparison is carried out, the audio frequency whose correlation coefficient is greater than the threshold value is continued to calculate the distance from the audio frequency of the corresponding time period in the audio library, and the channel number corresponding to the least distance is taken as the discriminant result, at the same time, By using the method of multi-packet data discrimination, two to three packets of data are taken for each channel watched by the user, so as to ensure the accuracy of the discrimination results. Finally, the proposed audio retrieval algorithm is evaluated to verify its effectiveness. As a new type of ratings survey system, the characteristics of the data taken in this paper are all based on the newly built system, and the data taken can meet the characteristics of small amount of data and high efficiency, thus reducing the complexity of the client-end equipment. Communications cost less and simplify the complexity of central data processing, and the communication module has multi-mode interface to meet the requirements of telephone transmission, wireless transmission and Internet transmission. Finally, the paper summarizes the work done in this paper, and points out the improvement in the future, as well as the future research direction.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TN912.3;TP391.3
，
本文编号：2436846
[Abstract]:Rating survey instrument is a kind of equipment invented after the traditional questionnaire survey. It is widely used in radio and television department and has high research value. The result of ratings survey is of great significance for TV stations to arrange programs better, advertisers to place advertisements in good time, and television groups to receive program information better. At present, the principle of rating survey equipment is diversified, the cost of equipment applied in radio and television department is high, and the reliability of equipment needs to be improved. In view of this background, this paper studies a new rating survey device based on audio features and its corresponding audio retrieval method. This paper first briefly introduces the relevant knowledge of ratings, ratings survey, etc., clarifies the main indicators of the ratings survey, the method and significance of the ratings survey, and then summarizes the main types of ratings survey equipment at present. At last, this paper puts forward the focus and content of this paper-the ratings survey system based on audio feature and the algorithm of ratings retrieval, and expounds the methods of audio signal processing and the main techniques of audio retrieval. The overall architecture of the ratings survey system based on audio features is composed of the audio data acquisition part at the user end, the data processing part at the central end and the communication module between the two. In this paper, the concrete functions of these three parts and the problems to be solved in their implementation are discussed, and the feasible implementation scheme is given. This paper focuses on the effective feature vector and audio retrieval methods based on the audio feature rating survey system in order to reduce the cost of the equipment and improve the availability and reliability of the equipment. The audio feature designed in this paper is an improved audio feature aiming at the problems existing in the audio acquisition of the actual TV viewing. The pass band division technology is adopted, and the pass bands sensitive to the high and low audio are eliminated. At the same time, The normalization method is used to avoid the influence of different volume on the acquired audio features, so that the audio features that are not affected by the volume and model of the TV set are obtained, and the audio features with good robustness are obtained. The reliability of audio characteristics is analyzed and studied by using MATLAB software, and the experimental results are given in the form of graph. The audio retrieval algorithm designed in this paper is carried out in two steps. First, the correlation calculation is carried out, and the correlation coefficients between the audio to be tested and the corresponding time period in the audio library are calculated, and the threshold decision method is adopted. The audio frequency whose correlation coefficient is less than the threshold value is eliminated; Then the Euclidean distance comparison is carried out, the audio frequency whose correlation coefficient is greater than the threshold value is continued to calculate the distance from the audio frequency of the corresponding time period in the audio library, and the channel number corresponding to the least distance is taken as the discriminant result, at the same time, By using the method of multi-packet data discrimination, two to three packets of data are taken for each channel watched by the user, so as to ensure the accuracy of the discrimination results. Finally, the proposed audio retrieval algorithm is evaluated to verify its effectiveness. As a new type of ratings survey system, the characteristics of the data taken in this paper are all based on the newly built system, and the data taken can meet the characteristics of small amount of data and high efficiency, thus reducing the complexity of the client-end equipment. Communications cost less and simplify the complexity of central data processing, and the communication module has multi-mode interface to meet the requirements of telephone transmission, wireless transmission and Internet transmission. Finally, the paper summarizes the work done in this paper, and points out the improvement in the future, as well as the future research direction.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TN912.3;TP391.3
，
本文编号：2436846
本文链接：https://www.wllwen.com/wenyilunwen/guanggaoshejilunwen/2436846.html
