摘要
对琵琶7种常用演奏指法进行了自动识别.首先,对声音信号声谱图根据其基音频率截取一部分并进行归一化处理;其后,根据归一化声谱图生成3种能明显体现不同指法特点的网格图;最后,从网格图中划分若干种不同的计算区域,用各区域的统计值作为特征.在基于机器学习的指法自动识别的实验中,对7类指法以及乐音类、泛音类指法的自动识别能够以很少的特征数量达到100%的识别率,实现了对琵琶指法的精确地自动识别.
Automatic recognition on seven typical fingering methods of Pipa is taken out.First,a selected part of Pipa spectrogram is normalized according to its fundamental frequency;then,3 types of grid diagrams are generated from the normalized spectrogram;finally,different computation zones are assigned to the grid diagrams for feature extraction with their statistical values.In the automatic recognition based on machine learning,accurate recognitions are achieved with accuracies up to 100%by using very few numbers of features.
作者
肖仲喆
张晶晶
刘弋嘉
XIAO Zhongzhe;ZHANG Jingjing;LIU Yijia(School of Optoelectronic Science and Engineering,Soochow University,Suzhou,Jiangsu 215006,China)
出处
《复旦学报（自然科学版）》
CAS
CSCD
北大核心
2020年第3期286-292,共7页
Journal of Fudan University：Natural Science
基金
国家自然科学基金(61906128,61802272)
上海市音乐声学艺术重点实验室合作项目(SKLMA-2019-04)。
关键词
琵琶
声谱图
网格图特征
Pipa
spectrogram
grid diagram features
