【作者】 王芳；

【导师】 刘齐跃；

【作者基本信息】 河北科技大学 ， 通信与信息系统， 2012， 硕士

【副题名】基于旋律的哼唱音乐检索

【摘要】 近年来，随着多媒体及网络技术的飞速发展，多媒体信息的数据量急剧的增多，人们已经不能满足于通过歌名、歌词、演唱者等信息进行的文本检索。特别是当人们只记得歌曲的某一旋律片段时，如何快速的通过音乐旋律来检索相关音乐就成为了现在急需要解决的问题。基于旋律的哼唱检索技术现已成为当今的研究热点之一，但直到现在它还没有被广泛的推广起来。这主要是因为，现在大多数的哼唱检索系统都是基于对MIDI音频文件的检索，并且对哼唱者及哼唱环境有着严格的要求。在现实生活中，人们接触到的都是MP3格式、WMA格式及WAV格式的音乐文件，MIDI格式的音乐文件少之又少，因此如何对MP3等格式的音频文件进行哼唱检索，并且哼唱条件适用于普通大众，这就是本文解决的问题。针对上述要解决的问题，本文主要做了如下几方面的工作：1)通过对声音特征参数的分析，从中选出决定乐曲主旋律特征的参数。这一特征参数要满足：不管什么人、用什么样的方式演唱，只要音调相同，特征参数就要相同。2)由于本文要检索的是带有背景音乐的音频文件，因此本文通过对常见音符音高值算法的分析研究，提出了一种新的音符音高值算法。这种算法不但计算量小，同时也降低了背景音乐对检索工作造成的困扰。3)通过对音阶与音高间存在的特定关系的分析，在常见的旋律匹配算法的基础上，针对它们各自的优缺点提出了一种改进的字符串匹配算法，进而提高了匹配检索的准确度。4)最后在上述分析的基础上，实现了在普通环境下对MP3、WAV等不同格式的音频文件的哼唱检索，并对多种不同的情况进行了仿真实验比对及结果分析。更多还原

【Abstract】 In recent years, with the rapid development of multimedia and network technology,multimedia information dramatically increased the amount of data, it is not enough topass information such as song name, the lyrics, the singer for text retrieval. Especiallywhen people only remember a melodic fragment of song, how quickly retrieve relatedmusic through the music of there was an urgent need to resolve the problem.Humming retrieval melody-based technology has now become one of the hot topicsof today, but until now it has not been widely promoted. This is mainly because, most ofthe QBH systems are currently based on the MIDI audio file retrieval, and has strictrequirements on the humming and humming environment. In real life, people areexposed to are MP3format and WAV format, WMA format music files, very little musicin MIDI format files, so how to humming to retrieve the audio file formats such as MP3,and humming conditions shall apply to the general public, this is the problem of thisarticle. Problems to be solved in view of the above, the article do the following work:1) Through to the acoustic feature parameter analysis, choose the decision keymelody characteristic parameters. This parameter to satisfy: no matter whose with whatkind of means to sing, as long as the same tone, characteristic parameter will be thesame.2) Because of this article is to be retrieved with the audio file for the backgroundmusic, so this article through the study of common notes pitch algorithm, proposed adifferent note pitch algorithm. This algorithm not only small amount of computation, butalso reduces the problem of background music to retrieve work.3) Through the analysis of the precise relationship between scales and pitch, melodymatch algorithm in common on the basis for their respective advantages anddisadvantages presented an improved algorithm for string matching, thereby improvingthe matching accuracy of retrieval.4) Finally, on the basis of the above analysis, implemented in the general context ofMP3, WAV and other audio files in different formats humming retrieval and simulationof many different experiments and results analysis.更多还原