MIDI哼唱检索算法研究+文献综述
时间:2017-06-25 19:27 来源:毕业论文 作者:毕业论文 点击:次
| 摘要音乐检索是在其他电子音乐库库中检索目标音乐的技术.音乐检索具有很强的应用背景在海量数据存储中,如果对每一首乐曲都通过人工做标注然后通过文本检索必然带来巨大的工作量当数据量越来越多时人工的注释强度加大人的工作量.但是对音频的感知,如音乐的旋律音调音质等难以用文字注释表达而清楚这些正是基于内容的音频检索需要研究和解决的问题. 对于一般人来说由于不具备专业的谱曲知识很难把自己感受到的旋律写下来而通过旋律抽取技术我们可以将旋律记录下来并可以通过电子和声的方法使用不同的乐器和音色来演奏,那么以后有音乐天赋的普通人也可以很好的使用方便查询音乐数据库中存储的数据.声音图像图形等音乐作为人类最直接的感知信息的模式是人们日常生活中不可缺少的部分在各个领域起着极大的作用由于经年来互联网带宽和速度的极大提高使得网络音乐资源成为极具商业价值和社会价值的资源但是面对庞大而分散的音乐资源如何随时检索到喜欢的音乐成为一个重大问题从数据库中利用各种可以表达音乐的手段检索到所需要的乐曲被称为音乐检索. 在数字音乐的研究应用中，大多数高层语义的音乐特征研究都以MIDI音乐文件库作为素材，其中主旋律音轨作为最重要的音乐信息，而从MIDI文件中获取主旋律信息具有很大的难度和误差。本论文提出一种应用随机森林的方法自动抽取MIDI文件中主旋律所在的音轨。首先根据信息熵理论定义了音轨特征的熵值，并通过实验判断该特征的熵值对区分MIDI主旋律音轨与伴奏音轨是有效的。然后由MIDI文件的音轨信息熵和一些其他重要特征组成特征向量，构建随机森林分类器抽取MIDI文件主旋律所在的音轨。根据实验可知，该模型有较高的准确率，对提取MIDI文件主旋律音轨有效。关键字：哼唱，MIDI音乐，检索，音轨10740 Music retrieved is in other electronic music library library in the retrieved target music of technology. music retrieved has is strong of application background in mass data storage in the, if on each first music are through artificial do callout then through text retrieved necessarily brings huge of workload Dang data volume increasingly multi artificial of comments strength increased people of workload. but on audio of perception, as music of melody tone sound quality, difficult to with text comments expression and clearly these is based on content of audio retrieved needs research and solution of problem. For most people, because they do not have professional knowledge of music is hard to write melodic feeling down we can tune recorded by Melody extraction technology and can be used by electronic and acoustic methods to a variety of musical instruments and sound playing, So yihou has music gift of ordinary people also can is good of using convenient query music database in the storage of data. sound image graphics, music as human most directly of perception information of mode is people everyday life in the not missing of part in all area up with great of role due to years to Internet bandwidth and speed of great improve makes network music resources became very with commercial value and social value of resources but face huge and dispersed of music resources how at any time retrieved to like of music became a major problem from database in the using various can expression music of means retrieved to by needs of music is called music retrieved. Study on application of digital music, most high level semantic characteristics of music with MIDI music file libraries as source material, in which melody track as the most important music information, obtained from the MIDI file and theme information with great difficulty and error. Paper presents a method of using random forest automatically pulls the melody MIDI file tracks. First track feature are defined based on information entropy theory of entropy, and experimentally determine the entropy to differentiate the characteristics of audio MIDI melody and accompaniment tracks are valid. And then by the MIDI file tracks information entropy and a number of other important features of eigenvectors, building a random forest classifier track where the main melody extraction MIDI file. According to experiments, which have a high accuracy rate, valid for the extraction of main melody MIDI file tracks. (责任编辑：qin) |