哼唱曲调识别与记谱的迭代改进算法  

An Improved Iterative Algorithm for Humming Tune Recognition and Notation

作　　者：杨岱锦 帅子恒 罗文博[1] YANG Dai-jin;SHUAI Zi-heng;LUO Wen-bo(School of Electronic Science and Engineering,University of Electronic Science and Technology of china,Chengdu610054)

机构地区：[1]电子科技大学电子科学与工程学院,成都610054

出　　处：《电子科技大学学报》2020年第2期228-234,共7页Journal of University of Electronic Science and Technology of China

基　　金：国家自然科学基金(51602039)。

摘　　要：哼唱记谱是音乐创作谱曲的重要方法与过程。该文分析了受多种环境因素影响的复杂哼唱音频基本特征;基于加窗傅里叶变换方法,以音符为单位对哼唱音频进行区域性的划分、定义和特征提取,提出了以相对振幅为依据快速提取基频的方法,设计出一种可变区域的傅里叶变换迭代算法。采用Python 3.6编程实现了上述迭代算法,自动获取哼唱音符的音高和音长并形成数字乐谱,实验测试准确率达到84.3%。上述结果表明,该算法能更加精确地识别哼唱曲调,为开发精准辅助作曲软件提供了一种可行的识别与记谱算法,具有较好的应用前景。Humming notation is an important method and process of composing music.Considering the complexity of humming audio and the influence of various environmental factors,this paper analyzes the basic characteristics of humming audio.On the basis of windowed Fourier Transform,the humming audio is regionally divided,defined and extracted according to the notes.A method of fast extraction of fundamental frequency is proposed based on relative amplitude of humming audio.And further a variable-region Fourier Transform iteration algorithm is designed and implemented programmatically by Python 3.6.This iteration algorithm can recognize humming melody more accurately,obtain the pitch and length of each note of humming,and automatically form a digital music score.The accuracy of the experimental test reached 84.3%.The achieved results show that the algorithm can identify humming tunes more accurately,thus it would be a feasible recognition and notation algorithm for developing composing-assisting software with good application prospects.

关 键 词：音频识别 基频提取 哼唱记谱 迭代算法 加窗傅里叶变换 

分 类 号：TP311.1[自动化与计算机技术—计算机软件与理论]