电视广告监管系统的设计和实现

发布时间：2018-04-30 01:20

本文选题：音频 + 语音匹配　；参考：《浙江大学》2006年硕士论文

【摘要】：随着经济的发展,电视广告成为社会生活中越来越重要的一部分,而其带来的社会问题也日渐显著,特别是虚假广告严重误导了消费者,坑害了广大人民,因此广告监管成为社会急需处理的问题。本文根据实际的项目要求设计了一套适合广告监管的系统,它主要可以分为视频采集、音频匹配和用户界面这几部分,并重点研究了适合电视广告监管的音频匹配算法。本文在总结语音识别和音乐检索技术的基础上,成功的实现了广告音频匹配算法。整个算法可以分成音频特征提取和音频模板匹配两大部分。在音频特征提取阶段,本文借鉴了音乐检索中主流的两种音频特征:音高和MFCC特征参数,并设计了一套实验来选择对广告音频有更强区分能力的特征。根据实验结果MFCC特征参数具有更好的识别能力,因此它将作为整个音频匹配算法中使用的音频特征参数。在音频模板匹配阶段,为了解决MFCC特征参数数据量过大、比较速度过慢以及广告音频中出现的多帧、丢帧等问题,本文应用了矢量量化技术、多级匹配算法和DTW算法。在矢量量化中,针对难以为整个广告样本库建立一个统一码本的情况,本文采用为单一广告分别建立码本的方法,并在量化时把不属于该广告空间的矢量统一划入另一区域来提高匹配时的精确度。在多级匹配算法中,采用了广告头比较来寻找广告开始帧的位置,用整个广告比较来确认广告头找出的开始帧位置是否正确。这种改进算法使匹配速度有了很大的提高。在确定了音频模板的匹配算法以后,本文又设计了一套实验来检测该算法的有效性,实验表明,在准确率没有明显下降的情况下,速度有了很大的提升。最后本文就整个项目的实现,进行了说明。
[Abstract]:With the development of economy , TV advertising has become a more and more important part of social life , and the social problems brought by it are becoming more and more obvious , especially the false advertisement seriously misled consumers , and harmed the people , so advertising supervision has become the problem of social emergency treatment .

According to the actual project requirement , a system suitable for advertisement supervision is designed , which can be divided into two parts : video acquisition , audio matching and user interface , and the audio matching algorithm suitable for TV commercial supervision is researched .

On the basis of summarizing the technology of speech recognition and music retrieval , this paper successfully realizes the advertisement audio matching algorithm . The whole algorithm can be divided into two parts : audio feature extraction and audio template matching .

At the stage of audio feature extraction , this paper draws on two kinds of audio features in music retrieval : pitch and MFCC feature parameters , and designs a set of experiments to select features that have stronger distinguishing ability for the advertisement audio . According to the experimental results , the MFCC feature parameters have better recognition capability , so it will be used as the audio feature parameter used in the whole audio matching algorithm .

In this paper , vector quantization , multilevel matching algorithm and DTW algorithm are applied in order to solve the problems of too large data volume , too slow speed of comparison and multi - frame and frame loss occurring in advertisement audio in the matching stage of audio template .

In the vector quantization , in the case of difficulty in establishing a uniform codebook for the entire advertisement sample library , a method for establishing a codebook for a single advertisement is adopted , and the accuracy of matching is improved when the vector which does not belong to the advertisement space is uniformly drawn into another area at the time of quantization .

in that multi - stage matching algorithm , the position of the advertisement start frame is searched by the comparison of the advertisement head , and the whole advertisement comparison is used to confirm whether the position of the start frame found by the advertisement head is correct .

After confirming the matching algorithm of the audio template , this paper designs a set of experiments to detect the effectiveness of the algorithm .

In the end , this paper describes the whole project .

【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2006
【分类号】：TP311.52

【相似文献】