有效的哼唱旋律基音提取方法研究与实现  

Research and implementation of effective pitch extraction method in singing melody

作　　者：蒋永平[1] 卢传泽[1] 于琪[1] 赵冲[1] 

机构地区：[1]广东工业大学信息工程学院,广州510006

出　　处：《计算机工程与应用》2014年第3期215-220,共6页Computer Engineering and Applications

基　　金：广州市番禺区2010年科技计划项目(No.2010-z-40-1)

摘　　要：为使用户可以通过哼唱音乐片段的方式,在较大规模乐曲库中能实时、准确、快速搜索到目标歌曲,结合多种音乐的乐理特点和音乐的数字化表示,提出考虑周期性测度改进传统端点检测方案,并综合线性预测分析(LPC)和相关函数与削波法相结合的方法进行旋律基音提取。实验表明利用LPC仅仅保留声门激励信号的特点,有效地减少了声道共振峰的影响,而采用削波法也一定程度减少噪声的干扰并使得计算相关函数的运算量大为减少,更显著的是考虑周期性测度的端点检测可以找到更精确的语音段,可以准确提取旋律并有效达到哼唱检索实时性要求。For searching the target song in real-time, accurately and quickly in a rather large music library by humming music clips, integrating music theory characteristics and digitized representation of a variety of music, this paper presents a method of melody pitch extraction by the period measure improving traditional endpoint detection and integrated approach of Linear Predictive Coding（LPC）and correlation function with clipping method. Experiments show that LPC can reserve glottal excitation signal characteristics and eliminate the influence of the sound tract impulse response and clipping method can reduce the noise interference and cut down calculated amount dramatically. More significantly, the method increases the veracity of extracting music clips by period measure improving traditional endpoint detection. As a result, the method meets the requirements of accuracy and real-time for extracting melody and target song.

关 键 词：哼唱检索 端点检测 基音提取 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]