基于MIDI哼唱检索算法的研究

发布时间：2018-07-17 01:36

【摘要】：随着多媒体技术的发展，以音频、视频和图像信息为主体的多媒体信息逐步取代了文本信息。基于文本标记的传统信息检索技术难以实现对多媒体信息的检索，如何有效、快速地检索多媒体信息，成为搜索引擎发展中迫切需要解决的问题。基于内容的音乐信息检索（Content Based Music Information Retrieval,CBMIR）是根据音乐的内在属性，，提取出音乐的特征向量（含节奏、旋律和音强等），构建音乐特征数据库；以乐谱、自然哼唱等形式提交检索项，将提取的特征向量与音乐特征数据库进行比较计算相似度，实现检索匹配。哼唱检索（QueryBy Humming, QBH）作为一种最直接、最自然的输入方式，自然成为基于内容检索技术的研究重点，并具有广泛的应用前景。哼唱信号处理、音乐特征数据库构建和检索匹配算法是研究的重点。研究哼唱检索信号处理流程，介绍了信号预处理、参数提取及基于能量和音高变化的音符分割算法，针对哼唱信号，重点分析信号特征，并依此提出了改进的音符分割算法；分析MIDI音乐文件结构和MIDI旋律信息提取算法，构建了音高差，音长差，音符时长，音符间隔的旋律表示方式，能够有效克服音符连音和音符间隔的问题；介绍4种常用的检索匹配算法，并针对动态时间规整算法，重点从哼唱时出现的音符连音哼唱及音符间隔等两个方面进行改进，根据音符连音时的特点及音符间隔对检索匹配位置的作用，改进了动态时间规整算法。在哼唱检索平台上测试了改进的音符分割算法和动态时间规整算法，实验结果表明改进算法的有效性。
[Abstract]:With the development of multimedia technology, multimedia information with audio, video and image information as the main body has gradually replaced text information. The traditional information retrieval technology based on text marking is difficult to realize the retrieval of multimedia information. How to retrieve multimedia information effectively and quickly has become an urgent problem in the development of search engine. Content based Music Information Retrieval (CBMIR) is to extract the music feature vectors (including rhythm, melody and strong tone etc.) according to the intrinsic attributes of music, to construct the music feature database, and to submit the retrieval items in the form of music score and natural humming. The extracted feature vector is compared with the music feature database and the similarity is calculated to achieve retrieval matching. As one of the most direct and natural input methods, QueryBy humming (QBH) has naturally become the focus of research on content-based retrieval technology and has wide application prospects. Hem signal processing, music feature database construction and retrieval matching algorithm are the focus of research. This paper studies the processing flow of humming retrieval signal, introduces signal preprocessing, parameter extraction and note segmentation algorithm based on energy and pitch changes, analyzes the characteristics of Hem signal, and puts forward an improved note segmentation algorithm. Based on the analysis of midi music file structure and midi melody information extraction algorithm, a melody representation method of pitch difference, note length difference and note interval is constructed, which can effectively overcome the problems of note concatenation and note spacing. This paper introduces four common retrieval and matching algorithms, and focuses on improving the dynamic time warping algorithm from two aspects: note concatenation and note spacing. According to the characteristics of note concatenation and the effect of note interval on searching matching position, the dynamic time warping algorithm is improved. The improved note segmentation algorithm and the dynamic time warping algorithm are tested on the humming retrieval platform. The experimental results show the effectiveness of the improved algorithm.
【学位授予单位】：江西师范大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.3

【参考文献】