【作者】 刘艳；

【导师】 郑贵滨；

【作者基本信息】 哈尔滨理工大学 ， 计算机应用技术， 2008， 硕士

【摘要】 哼唱音乐检索是一种有效的基于内容的音乐检索方式,它具有自然、方便的特点,有极其广阔的应用前景和重要的研究价值,是近年来国内外研究的热点课题。本文针对MIDI音乐检索,主要研究MIDI数据库的建立、哼唱特征提取和旋律匹配算法三方面内容。本论文在数据库的建立中使用MIDI格式文件作为提取音乐旋律的数据源。在分析了MIDI音乐描述格式的基础上,采用了基于主旋律音轨特点的MIDI音乐主旋律提取算法,并对其进行了实验验证,虽然实验结果表明,该算法的性能还可以,但为了达到最佳的效果,本文最终采取的还是人工监听的方式,手工标注每首MIDI音乐的主旋律音轨。在哼唱特征提取中,为了解决音符分割、音符识别难度大的问题,提出了一种基于两级神经网络的哼唱特征提取方法。第一级采用BP神经网络实现哼唱音符分割,得到独立音符;第二级采用RBF神经网络识别分割出的各个音符,获得音符的MIDI音高值。实验结果表明,该方法能较好地完成哼唱特征的提取,适合于实际哼唱检索系统对连续哼唱的要求。在旋律匹配算法中,利用旋律的音符序列具有时间上的可伸缩性和音高上的可平移性,先把两段旋律在时间轴上线性延展到相同的长度,并在一定的误差范围内对齐发声时刻接近的音符,考察旋律在节奏上的相似性之后,继续比较两段等长旋律在每个时间点上音高的差距。最后,综合考虑节奏和音高两方面的相似程度,给出匹配得分,将得分最高若干乐曲作为检索结果。更多还原

【Abstract】 Query by humming is an effective content-based music retrieval. It is a natural and convenient music retrieval method and it has great value in many application areas. Query by humming is a hot research topic at home and abroad in recent years. This paper studies the establishment of a database of MIDI, humming feature extraction and melody matching algorithm.In the establishment of the database, this paper used the MIDI format as data sources that music melody is extracted from. MIDI music theme extraction algorithm is used on the basis of MIDI music’s format and verified by the experiment. Although the experimental results show that the performance of the algorithm can be, but in order to achieve the best results, the paper eventually choices the artificial monitor way and receives the theme’s tracks of each first MIDI by manual tagging.In the Feature Extraction, a two-layer neural network for humming feature extraction is proposed for note segmentation and note recognition. The first-layer BP network is used to divide the humming data into the independent notes; the second-layer RBF network is used to receive the MIDI pitch of the independent note. The experimental results show that this method can extract the humming features with high accuracy, and it is suitable to the query by humming system.In the Melody matching algorithm, using the scalability of time and the translational of pitch in the notes sequence of the melody, first, two melodies are extended linearly to the same length in the time frame and aligns the notes whose onset are close in a certain time range. Then, the two melodies of same length are compared to the gap between the pitch at each time point, after the similarity of melody is studied in the rhythm. Finally, the paper gives the matching scores by considering the similarity both the rhythm and pitch and the pieces of the highest score will be retrieved as a result.更多还原