算法音乐诱发脑电的多特征分析

Multi-feature Comparison of Algorithmic Music’s Influence on EEG

为了探究大脑对算法作曲的低质量音乐与人为编曲的高质量音乐之间的反应差异性,采用长短时记忆网络(Long short-term memory,LSTM)生成不同可听度的算法音乐,从生成的算法音乐中选取部分音乐按综合评价分数划分为5个级别,并将流行音乐旋律节选作为第6级,组合进行范式实验。使用微分熵、功率谱密度、Hjorth参数和其他频域分析的特征提取方法对实验脑电信号进行特征提取、数据分析和显著性判断,对比在听不同可听度音乐时受试者脑电信号特征的差异性。进一步分析不同性别和不同音乐专业背景的受试者在听到同样音乐时脑电特征的差异性。实验结果表明功率谱密度(Power Spectral Density,PSD)反映脑电信号差异性效果最好。

In order to explore the difference between the brain’s response to low-quality music composed by algorithms and high-quality music composed by humans,Long short-term memory(LSTM)was used to generate algorithmic music with different audibility.The selected part of the algorithmic music was divided into 5 levels according to the comprehensive evaluation scores,and the popular music melody excerpts were regarded as the 6th level,used as the music material for paradigm experiments.Feature extraction methods including differential entropy,power spectral density,Hjorth parameters and other frequency domain analysis,were used to the feature extracting,the data analysing and the significance judgement of experimental electroencephalogram(EEG)signals,which were carried out to compare the difference of EEG signals when subjects were listening to music with different audibility.The differences of EEG characteristics of subjects with different gender and music professional background when listening to the same music were further analysed.The experimental results show that the power spectral density(PSD)is the best feature extraction method for comparing the difference of EEG signals.