Research On Affective Computing Methods Based On Auditory Cognitive Principles

Emotion is a basic component of social interactions and cognitive activities of human beings.With the help of effective emotional interactions,the user experience of humancomputer interaction(HCI)will be greatly enhanced.Auditory and vision are the most two important sources of human's emotion.Due to the characteristics of the auditory system structure,the auditory emotional cognitive process is more complicated.The basic problems of the auditory emotion mechanism,evoking methods,influencing factors and cognitive principles are still unclear.How to make the computer have the advanced emotional processing ability and realize the natural HCI technology is an urgent problem in the field of artificial intelligence(AI).At present,most of emotional recognizing methods are based on statistical or machine learning methods? Lacking of cognitive principles greatly limits the development of auditory emotion computing technology.The most effective and direct research method is to start from the information processing mechanism of the human brain emotional system.Therefore,under the challenges of auditory emotion processing,this paper focuses on the auditory emotional cognitive principles,the analysis methods of emotional EEG signals,the EEG based emotional state recognition,brain-inspired music emotion recognition and other core issues.The main research contents include:(1)The precise cognitive experiment paradigm is the premise of the analysis of emotional cognition.Aiming at the problem of lacking long-term auditory emotion cognitive experiment paradigm and cognitive principle analysis method,this paper proposes different semantic levels as sound,music and speech experimental methods: improved N-back experimental paradigm for short-term sound and speech;a three-stage emotional evoked experimental paradigm for long-term music.Secondly,for the problem of analyzing the long-term EEG signals,a long-term event-related potentials(ERP)analysis method and a long-term EEG spectrum analysis method are proposed,which realize a dynamic description of the emotional change process.Finally,the findings of the auditory emotional cognition principles and their application scenarios are summarized.(2)Aiming at the problem that the traditional feature extraction methods can't eliminate the individual difference of EEG signals,an adaptive EEG analysis method based on emotional EEG spectral difference(EESD)and Optimized CQT Transform is proposed.How to accurately and efficiently extract EEG features plays an important role in the analysis of emotional EEG.Therefore,based on the cognitive principles of EEG bands,an emotional individual difference calculation index based on emotional EEG spectral response difference EESD is proposed to realize the calculation of individual differences.At the same time,through improved CQT transformation,EEG The spectral difference of the signal is accurately expressed.Secondly,the CQT transformation is optimized by EESD to eliminate the individual difference of EEG signals.Experiments have been carried out on the real-acquired music-modulated emotional EEG data and public datasets.The accuracy of this method is about 3% and 4% higher than the baseline using the Welch method,which can better reveal the individual characteristics of auditory-evoked emotional EEG spectrum distribution.(3)Aiming at the problem of feature extraction of emotional EEG signals,a Common Spatial Pattern(CSP)analysis method based on cognitive topology constraint matrix(TCM)is proposed.The EEG signals with a large number of electrodes contain a large amount of redundant information,which will increase the computational complexity and reduce the recognition rate.Aiming at the problem of CSP feature optimization,we propose a CSP feature analysis method based on convolutional neural network(CNN),which uses the CNN network to learn the feature matrix,analyzes the obtained full-connection layer weight matrix,defines the feature selecting criteria and obtains an efficient EEG feature set.Secondly,the cognitive principle is introduced into the CSP calculation method,and the TCM based CSP calculation method is proposed.According to the distribution principle of human brain's emotional cognition brain region,the emotional topological constraint matrix TCM is proposed to extract emotion-related EEG features,which narrows the gap between features and emotion categories.Through the experiments on the public datasets,the output pattern can obtain the best classification recognition accuracy.(4)Aiming at the lack of the guidance of human emotion cognition principle and computational model,a music affective computing method based on brain inspired ?distribution is proposed.Firstly,based on the analysis of auditory characteristics and emotional response,an emotional analysis method based on ? distribution is proposed.Secondly,considering the relative relationship between the sample and the emotion,a representational dissimilarity matrix(RDM)based emotional feature analysis method is proposed to further analyze the emotional features.On these basis of emotion-related features,a method of emotion prediction based on Lasso regression is proposed to realize the prediction and recognition of music emotions.Finally,the overall structure of the auditory emotional state prediction system is given.In this paper,the key issues of auditory brain cognitive principle and emotional computing method are studied in detail,focusing on the analysis of auditory emotional brain cognitive principle,brain inspired emotional EEG signal feature extraction,emotional state recognition.The core issues of knowing the regular audio emotion recognition have laid a solid foundation for various cognitive signal processing.There are some enlightening significance for related fields that using the cognitive principles of as the guidance of affective computing recognition,and using the calculation results as the proof of brain science.