基于内容的音频检索的研究与实现
发布时间：2018-08-07 12:18
【摘要】： 音频信息是一类非常重要的多媒体信息。随着人们每次能够处理的音频信息量越来越大、音频信息的种类越来越繁多,要从海量的信息中迅速有效地检索出所需要的信息就变得越来越重要。然而,音频信息检索的研究一直没有得到足够的重视。尤其是基于内容的多模板的实时音频检索更是很少被人提及。 本文研究的是基于内容的音频检索。研究的切入点是电台广告,研究的目标是要在众多的音频电台节目中快速、有效的检出指定的几个电台广告,通过监播确定其实际播出的时间、终止时间以及播出次数等。 本文根据电台信号的实时性和连续性的特点,给出了监播系统的整体方案设计。将实时广告监播过程分为两个步骤:广告头定位和检出处理。广告头定位的目的是实时地判断出当前电台节目是否可能是待监测广告,如果是则进一步判断可能是哪几个广告,以便为广告的检出做一些指导性的工作。检出处理需要完成的工作则是准确判断出定位后的电台节目究竟是不是待监测广告,如果是则要准确指出是哪个广告。在广告定位前,对音频特征进行了有效的选取,在音频检索中以便能更快速的检出广告。运用音频特征对广告头片段运用门限法进行检索,并对广告头进行定位,从定位点截取广告的长度,作为广告待测模板。本文为检出处理采用了两种方案,便于实时处理的动态时间规整和聚类的矢量量化技术。并通过大量的实验对这些方法的可行性进行了验证和比较。并取得了较好的效果。
[Abstract]:Audio information is a kind of very important multimedia information. With the increasing amount of audio information that people can deal with each time, the variety of audio information becomes more and more diverse, so it is more and more important to retrieve the needed information quickly and effectively from the massive information. However, the research of audio information retrieval has not been paid enough attention to. In particular, content-based multi-template real-time audio retrieval is rarely mentioned. This paper focuses on content-based audio retrieval. The starting point of the research is radio advertising. The goal of the research is to detect several radio advertisements quickly and effectively in many audio radio programs, and to determine the actual broadcast time, termination time and broadcasting times through monitoring and broadcasting. According to the characteristics of real-time and continuity of radio signal, this paper gives the overall scheme design of monitoring and broadcasting system. The real-time advertising monitoring and broadcasting process is divided into two steps: ad header positioning and check-out processing. The purpose of ad header positioning is to determine whether the current radio program is likely to be an ad to be monitored in real time, and if so, to further determine which advertisements may be in order to do some guiding work for the detection of advertisements. What needs to be done is to find out exactly whether the radio program is to be monitored, and if so, which one. In order to detect advertisements more quickly, audio features are selected effectively before advertising positioning. Using audio features to search the advertising header using threshold method and locating the advertising head the length of the advertisement is intercepted from the location point as the template for advertising to be tested. In this paper, two methods are used for detection processing, such as dynamic time warping and vector quantization for real time processing. The feasibility of these methods is verified and compared by a large number of experiments. Good results have been obtained.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2006
【分类号】：TP391.3
本文编号：2170005
[Abstract]:Audio information is a kind of very important multimedia information. With the increasing amount of audio information that people can deal with each time, the variety of audio information becomes more and more diverse, so it is more and more important to retrieve the needed information quickly and effectively from the massive information. However, the research of audio information retrieval has not been paid enough attention to. In particular, content-based multi-template real-time audio retrieval is rarely mentioned. This paper focuses on content-based audio retrieval. The starting point of the research is radio advertising. The goal of the research is to detect several radio advertisements quickly and effectively in many audio radio programs, and to determine the actual broadcast time, termination time and broadcasting times through monitoring and broadcasting. According to the characteristics of real-time and continuity of radio signal, this paper gives the overall scheme design of monitoring and broadcasting system. The real-time advertising monitoring and broadcasting process is divided into two steps: ad header positioning and check-out processing. The purpose of ad header positioning is to determine whether the current radio program is likely to be an ad to be monitored in real time, and if so, to further determine which advertisements may be in order to do some guiding work for the detection of advertisements. What needs to be done is to find out exactly whether the radio program is to be monitored, and if so, which one. In order to detect advertisements more quickly, audio features are selected effectively before advertising positioning. Using audio features to search the advertising header using threshold method and locating the advertising head the length of the advertisement is intercepted from the location point as the template for advertising to be tested. In this paper, two methods are used for detection processing, such as dynamic time warping and vector quantization for real time processing. The feasibility of these methods is verified and compared by a large number of experiments. Good results have been obtained.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2006
【分类号】：TP391.3
【引证文献】
相关博士学位论文 前1条
1 何新;基于内容的音频信息分类检索技术研究[D];南京理工大学;2007年
相关硕士学位论文 前2条
1 付涛;基于DSP的音频信号快速评测系统的硬件设计[D];哈尔滨工程大学;2011年
2 王亚平;基于Android的中文语音短信应用设计[D];内蒙古科技大学;2012年
，本文编号：2170005
本文链接：https://www.wllwen.com/wenyilunwen/guanggaoshejilunwen/2170005.html
