钢琴弹奏乐曲识别算法研究及其APP设计与实现

发布时间：2018-05-04 00:42

本文选题：乐曲识别 + 音符检测　；参考：《南京理工大学》2017年硕士论文

【摘要】：乐曲自动识别是新兴的交叉学科,在音乐检索领域和音乐自动谱曲技术中有很重要的应用价值。目前的乐曲识别研究热点主要集中于单音符的识别,并且在识别精度、抗噪性能等方面存在着一定局限性。本文针对此问题,在深入了解音乐体系的基础上,围绕钢琴乐曲识别中的诸多关键技术,如多音符起止点检测、帧基频精确提取、音符基频准确计算等展开探讨,为包含连续音符弹奏的钢琴乐曲识别提出了解决方案,主要工作如下:一是基于单门限能量差法实现音乐段与噪声段的分割,然后对音乐段基于短时能量差法进行音符起止点检测。上述方法利用了钢琴的音乐特性来识别能量跳变点,能够有效提高音符起止点检测准确率,避免漏判、错判的情况。二是在研究自相关法、倒谱法与短时幅度差法等方法的基础上,提出一种乐曲基频提取的改进方法,能够突出帧样本周期位置的峰值特性,从而避免了半频、倍频的影响,有效提高基频提取的精度。三是在分析比较音乐信号波形起伏特性及音符帧样本数据处理方法的基础上,改进了基频计算方法,该方法赋予音符中间帧更高的权系数,达到较传统方法更高的精度和容错性。最后,综合以上算法,搭建了一个基于安卓移动终端+服务器的乐曲识别应用系统,对算法进行测试,验证了算法的可行性和效率。
[Abstract]:Automatic music recognition is a new interdisciplinary subject, which has an important application value in the field of music retrieval and music automatic composing. The current research focus on music recognition is mainly focused on the recognition of single notes, and there are some limitations in recognition accuracy and noise resistance. On the basis of the music system, a number of key techniques in the recognition of piano music, such as the detection of multi note starting point and stop point, the accurate base frequency extraction of the frame and the accurate calculation of the basic frequency of the notes, are discussed. The main work is as follows: one is to realize the music segment based on the single threshold energy difference method. This method uses the musical characteristics of the piano to identify the energy jump points, which can effectively improve the accuracy of the detection of the starting and stop points, avoid the missing and misjudged cases. Two is in the study of autocorrelation, cepstrum and short-time amplitude difference. On the basis of the method, an improved method of fundamental frequency extraction of music is proposed, which can highlight the peak characteristic of the period position of the frame sample, thus avoid the influence of half frequency and frequency doubling, effectively improve the precision of the base frequency extraction. Three, on the basis of analyzing and comparing the wave characteristics of the music signal and the processing method of the sound frame sample data, the base of the analysis and comparison is improved. The method of frequency calculation gives a higher weight coefficient in the middle frame of the note, and achieves higher accuracy and fault tolerance than the traditional method. Finally, a music recognition application system based on Android mobile terminal + server is built, and the algorithm is tested, and the feasibility and efficiency of the algorithm are verified.

【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN912.3;TP311.56

【相似文献】