面向情感分析的古筝乐曲声谱图信息主干提取

Extraction of Key Information from Guzheng Music Spectrogram Aiming Mood Analysis

古筝是一种古老的中国传统民族乐器,为一弦一音的弹拨乐器.情感作为古筝演奏中的重要表达成分,受到各种演奏技法的影响.本文分析了各种不同演奏技法所产生的声音信号的频谱以及声谱图特点,并提出了一种基于声谱图有限局部峰值的乐曲信息主干提取方法.乐曲信息主干提取分为时域提取与频域提取两个步骤.首先,义甲触弦瞬间,会出现短暂的非谐波性质的碰撞声,与后续琴弦稳定振动期间的声谱图横条纹形成对比,通过对声谱图进行时域一阶差分,并根据设定阈值二值化,能够准确获得各音符的触弦时间点;其次,在每一音符时间范围内,对声谱图进频域一阶差分,获取各帧频域峰值位置,并对各峰值根据幅值进行排序后保留有限个峰值坐标,同时根据前后音符对比,去除因前一音符声音的延续造成的对后一音符声谱图的影响.通过乐曲声谱图主干提取,能够有效地体现出组成乐曲的各音符的主要信息,为进一步深入分析古筝曲的情绪提供基础.

Guzheng is a kind of Chinese traditional instrument with strings.As one of the important information in Guzheng music,mood is influenced by various playing techniques.Spectrogram of sound signal produced in different playing techniques are analyzed.A key information extraction method based on spectrum local peaks with limited number is also proposed.The extraction process is divided into two stages in time domain and frequency domain respectively.In the first stage,a very short non-harmonic sound will occur instantaneously when the mock nails touching the string,which causes different patterns in spectrogram in comparison with the later stable vibration of strings.First order difference is performed to the spectrogram,and binarized according to proper threshold,to get the stimulation time of each note.In the second stage,first order difference is performed within the duration of each note,to get the position of spectrum peaks in short frames.The spectrum peaks are sorted according to their amplitudes,and only limited numbers of large peaks are retained.Furthermore,the influence of the previous notes could be deleted via comparison of successive notes.This extraction can effectively present the main information of each note in Guzheng music,and provide the basis of further analysis of Guzheng music mood.