基于自相关函数的钢琴乐音改进识别算法  

Improved Piano Audio Recognition Algorithm Based on Autocorrelation Function

作　　者：刘莹[1,2] 赵彤洲[1,2] 江逸琪 柴悦 李翔 LIU Ying;ZHAO Tongzhou;JIANG Yiqi;CHAI Yue;LI Xiang(Hubei Key Laboratory of Intelligent Robot（Wuhan Institute of Technology）, Wuhan 430205, China;School of Computer Science ＆ Engineering, Wuhan Institute of Technology, Wuhan 430205, China;Wuhan Tianyu Chengdu Westone Information Industry Inc., Wuhan 430223, China)

机构地区：[1]智能机器人湖北省重点实验室(武汉工程大学),湖北武汉430205 [2]武汉工程大学计算机科学与工程学院,湖北武汉430205 [3]武汉天喻信息产业股份有限公司,湖北武汉430223

出　　处：《武汉工程大学学报》2018年第2期208-213,共6页Journal of Wuhan Institute of Technology

基　　金：国家自然科学基金(61103136);武汉工程大学研究生创新基金(CX2017076)

摘　　要：在传统三电平削波结合自相关函数识别算法的基础上,经过准确的音频分割后,提出了帧移法提取乐音基音信号。该算法能在更精细尺度上寻找最大自相关函数,进而准确定位基音位置,较好地解决了传统算法中当乐音节奏较快时,无法区分半频和倍频对基音的影响,从而导致的识别率低的问题。实验表明,本算法对于节奏快慢不同的钢琴乐音的平均识别率约为83.0%,且快节奏乐音的识别率较传统算法高出20.2%,因此该方法对乐音识别尤其对快节奏乐音识别有显著效果。Combined with traditional three-level center clipping method and autocorrelation function recognition algorithm, an improved frame-shift algorithm to extract precisely the pitch signal was presented, which could search the maximum autocorrelation function at a finer scale to accurately locate the pitch position after accurate audio segmentation. This algorithm solved the problem that the traditional algorithms could not distinguish the influence of half-frequency and double-frequency on the pitch with fast rhythm, which degraded the recognition rate. Experiments showed that the improved algorithm had an average recognition rate of 83.0% for piano music with different rhythms, and the recognition rate with fast-paced music was 20.2% higher than that of traditional method. Therefore, the proposed algorithm has a significant improvement on music recognition, especially for fast-tempo music.

关 键 词：基音周期 三电平中心削波 自相关函数 帧移 乐音识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]