Current research on music mood has primarily focused on the field of emotion recognition, with relatively few studies dedicated to the methods of mood representation and absence of localized music mood models. Accordingly, this paper designs a music mood model based on neural signals of electroencephalogram (EEG), electrocardiogram (ECG) and electromyogram (EMG), and explores the representation method of music mood. Its core methodologies are as follows: feature values such as average energy, amplitude difference and sample entropy are extracted from the EEG, ECG and EMG of subjects, while key features are selected via correlation analysis and random forest. Based on an applicability evaluation of mapping modes between key features and subjective ratings, regression terms are filtered using LASSO regression, and three formulas to predict Pleasure (P), Arousal (A) and Dominance (D) values are determined by comprehensive comparison of feature retention and performance metrics. Finally, test results using a self-constructed dataset of 40 testing samples indicate that the single prediction bias are mostly within 10%, for instance, the prediction bias of 88% of P samples, 93% of A samples and 90% of D samples are less than ±5%. The 3-D convex hull constructed from 295 training samples indicate that the overall predictions of music mood model have significantly smaller fluctuation than subjective ratings. Thus, the method of mood representation based on neural signals has higher credibility.
音乐是由声音通过媒介按一定规律组织而成的,它被人类听觉器官接收、在大脑皮层诱发生物电刺激,继而产生诸多感性的心理变化.《礼记·乐记》中记载:“凡音之起,由人心生也”,故音乐被认为是表达情感、塑造意象的艺术形式和文化活动.20世纪50年代后,一门将音乐与科技紧密结合的交叉学科(即音乐科技/计算机音乐)悄然兴起.作为音乐科技的重要方向,“音乐信息检索”(Music Information Retrieval, MIR)专注于使用数字技术理解和分析音频内容(尤指音乐).[1]对音乐情感的分析、表示、识别是MIR子课题“音乐高层语义分析”的研究内容之一,在音乐制作、标注、推荐、理疗等场景有着广泛的应用[2].
构建模型是从情感理论走向情感计算的关键步骤.心理学界主要有两种方式——分类概念化和维度概念化[9].分类概念化从范畴观的角度对情感进行划分,国内称为离散模型.音乐情感方向的离散模型始于1936年Kate Hevner归纳的8类别、66个情感形容词[10].维度概念化是利用向量值表示情感在二维或三维空间中的特定位置,国内称为维度模型.1974年,Albert Mehrabian和James A. Russell解释了基础情感的三个维度——愉悦度(Pleasure,以下简称P)、唤醒度(Arousal,以下简称A)和支配度(Dominance,以下简称D)[11].1980年,James A. Russell将愉悦度改为效价度(Valence,以下简称V),构建了二维象限的情感模型,其横轴表示V、纵轴表示A.离散与维度模型分别对应了计算科学中的分类法和回归法.前者易于理解、可解释性强,但碍于语言、文化间差异和文字表述的模糊性,通用性和量化能力较弱.后者所描述连续过渡、动态转化的情感状态,更贴近人类情感的实际情况,在情感计算领域更具可行性.
DennisT A, SolomonB. Frontal EEG and emotion regulation: Electrocortical activity in response to emotional film clips is associated with reduced mood induction and attention interference effects[J]. Biological Psychology, 2010, 85(3): 456-464.
[25]
WaghK P, VasanthK. Performance evaluation of multi-channel electroencephalogram signal (EEG) based time frequency analysis for human emotion recognition[J]. Biomedical Signal Processing and Control, 2022, 78: 103966.
[26]
ErgoLAB Dry-EEG干电极脑电仪-津发科技[EB/OL]. [2025-10-31].
[27]
ErgoLAB HUB穿戴式多模态生理记录仪-津发科技[EB/OL]. [2025-10-31].
[28]
BradleyM M, LangP J. Affective reactions to acoustic stimuli[J]. Psychophysiology, 2000, 37(2): 204-215.
PopivanovD, MinevaA. Testing procedures for non-stationarity and non-linearity in physiological signals[J]. Mathematical Biosciences, 1999, 157(1-2): 303-320.
[34]
HuangN E, ShenZ, LongS R, et al. The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis[J]. Proceedings of the Royal Society of London Series A: Mathematical, Physical and Engineering Sciences, 1998, 454(1971): 903-995.
[35]
DragomiretskiyK, ZossoD. Variational mode decomposition[J]. IEEE Transactions on Signal Processing, 2014, 62(3): 531-544.
[36]
LiX, YanZ, GongP, et al. Research progress on emotion recognition based on electroencephalogram signals[J]. Biomedical Signal Processing and Control, 2026, 113: 109188.