海量音频指纹数据的存储与检索研究

发布时间：2018-04-19 01:05

本文选题：音频指纹 + 海量数据　；参考：《天津大学》2014年硕士论文

【摘要】：随着大数据时代的到来，尤其是包括图像、音频、视频在内的海量多媒体数据，这些数据亟需被有效地管理起来，并为广大用户提供方便、快捷的检索方式。随着模式识别、机器学习、云计算技术的发展，基于内容的多媒体检索技术应运而生，这种技术的出现使得信息检索不再依赖于数据的标签和关键字，而且搜索结果更为准确，搜索方式更为便捷。音频数据作为多媒体中重要组成部分，其数据规模也在迅速膨胀，，人们面临的问题不再是缺少多媒体信息，而是如何在海量的数据中找到自己所需要的信息。如何快速有效的检索海量音频成为当前学术界和工业界信息检索研究领域的一个重要课题。音频指纹检索技术是基于音频内容的信息检索方式，通过对未知音频片段提取名为音频指纹的数字特征，然后在事先准备的海量音频指纹数据库中进行音频指纹的搜索与相似度计算，获得音频详细信息的方法。这种方法解决了传统基于文本关键字的搜索音频存在的文本标注不全、错误等问题，同时解决了用户不知道关键词时无从下手搜索的困难。音频指纹提取与匹配算法已经在实验室中取得了丰硕的成果，并在部分产品中得到了应用，但所处理的数据集规模相对较小。在应用到大规模数据集时会遇到性能瓶颈，以及并发性、扩展性等问题。本文在对音频指纹提取与匹配算法的深入研究基础上，对海量音频指纹数据的存储与检索进行了设计、实现及优化。首先提出了基于哈希结构的音频指纹存储结构，然后进一步提出了两种分布式哈希解决方案，并通过实验证明了所设计的方法的有效性。在此基础上，本文又提出了一种海量音频指纹数据的序列化分布式存储方案，并再一次通过实验证明了其有效性。本文所设计的存储结构和分布式存储检索方案具有多级并发、高性能、可容错、易扩展等特点，对于构建海量音频指纹检索系统具有实际价值，对于推进音频指纹检索技术在社会中的应用具有重要意义。
[Abstract]:With the arrival of big data era, especially the mass of multimedia data including image, audio and video, these data need to be effectively managed, and provide convenient and fast retrieval methods for the majority of users.With the development of pattern recognition, machine learning and cloud computing technology, content-based multimedia retrieval technology emerges as the times require, which makes information retrieval no longer rely on data tags and keywords, and the search results are more accurate.Search is more convenient.Audio data as an important part of multimedia, its data scale is expanding rapidly, people are faced with the problem is no longer lack of multimedia information, but how to find their own information in the mass of data.How to retrieve massive audio quickly and effectively has become an important topic in the field of information retrieval in academia and industry.Audio fingerprint retrieval is an information retrieval method based on audio content.Then, the search and similarity calculation of audio fingerprint are carried out in the massive audio fingerprint database prepared in advance, and the method of obtaining audio detail information is presented.This method solves the problems of incomplete text tagging and error in traditional text keyword based search audio, and solves the difficulty of searching when users do not know key words.The algorithm of audio fingerprint extraction and matching has achieved a lot in the laboratory and has been applied in some products, but the scale of the data set is relatively small.Performance bottlenecks, concurrency and expansibility are encountered when applying to large data sets.Based on the deep research of audio fingerprint extraction and matching algorithm, this paper designs, implements and optimizes the storage and retrieval of massive audio fingerprint data.Firstly, the audio fingerprint storage structure based on hash structure is proposed, and then two distributed hash solutions are proposed. The experimental results show that the proposed method is effective.On this basis, a serialized distributed storage scheme for massive audio fingerprint data is proposed, and its effectiveness is proved by experiments again.The storage structure and the distributed storage retrieval scheme designed in this paper have the characteristics of multilevel concurrency, high performance, fault-tolerant, easy to extend, etc., which have practical value for constructing massive audio fingerprint retrieval system.It is of great significance to promote the application of audio fingerprint retrieval technology in society.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP333;TP391.3

【参考文献】