Emotion Cognition Research Based On The Analysis Of EEG And ERP Signals

Electroencephalograph(EEG)is involved with the electrical signals recording the brain activities of human beings;while event-related potential(ERP/ERPs)is the electroencephalograph which is induced by the specific stimulus,and its acquisition usually requires multiple EEG signals of the same event to be superimposed.To conduct characteristic analysis by using cognitive psychology and signal processing towards induced EEG signals of emotional speech and non-verbal sounds in nature can obtain deeper and more comprehensive emotion cognition information compared that by simply using cognitive psychology and signal processing.Hence,the former can provide the theoretic basis for the deeper research on man-machine interaction and offer valuable references for doctor diagnosis and patient handling.Therefore,the emotion recognition research based on the analysis of EEG and ERP signals has great theoretic significance and utility value.The research direction of this dissertation belongs to the cross subject of brain cognition and speech information processing.In the dissertation,the author makes comprehensive exploration and tries to solve the relevant research problems of emotional speeches and sounds by analyzing and processing the emotional EEG,as well as emotional speech and emotional non-speech signals based on cognitive psychology and signal processing.First of all,in order to collect effective emotional EEG signals,the solutions of screening experimental data and data pretreatment methods were provided.Second,according to the psychological cognition method,the questions were put forward that whether using different emotional speech acoustics parameters(fundamental frequency and duration)will affect the process of emotional cognition and whether speech and non-speech can influence the emotional cognition process.To address these questions,the experiments were designed on some assumptions and some predictions were given on the basis of the experiment results.Thirdly,based on cognitive psychology methods can not effectively identify the EEG signals,this dissertation proposed an improved compressed sensing for speech and EEG signals in the reconstruction of denoising,which can be used in the effective identification of single-trial ERP signals.Finally,from the prospective of psychological cognition and signal processing,the study content of this dissertation was summarized and evaluated.The main innovative points and contributions of the dissertation are shown as follows:(1)The screening and processing of emotion-inducing materials.Based on the fact that the existing TYUT2.0 emotion speech database cannot meet the experimental requirements,the author of this dissertation complemented and improved the emotional type(increased neutral speech)and quantity of TYUT2.0 emotional speech database,and established the TYUT2.1 non-speech emotional database.In addition,in order to facilitate the superposition of ERP waves,it has been presented to adopt terminal test approach to pre-process signals,by which,the soundless starting,wording space and duration of the first word of signals have been processed.According to the experiment result,it has been discovered that the components of ERP wave obtained after using the preprocessing method are more significant.(2)The influence of Chinese emotional speech‘s acoustic parameters on ERP components.Based on the question that the differences of Chinese emotional speeches in duration and average fundamental frequency on ERP components,the experimental exploration and research were conducted.By analyzing the components of N100,P200 and N300,the deduction is conducted towards research hypotheses.In terms of the research on duration,the experimental materials were divided into the ones with short duration(0.50-1.00 s),medium duration(1.50-2.00 s)and long duration(2.50-3.00 s).Besides,four emotions,i.e.,sadness,anger,happiness and surprise,are adopted for ERP analysis and research.It has been found out in the experimental result that when the duration of the induced material is short,it is easier to observe the differences among emotions by the amplitude of P200.In terms of the research on average fundamental frequency,the experimental materials are divided into the group that happiness is larger than sadness and the group that the sadness is larger than happiness.Besides,three emotions,i.e.,sadness,happiness and neutrality are adopted for ERP analysis.The experimental result shows that the amplitude of N100 components is influenced by the fundamental frequency of sounds.However,as for the components of P200 and N300,the amplitude does not totally rely on fundamental frequency.It is speculated that the amplitude is related to other sensational information of the speech.At last,by the experiment on the two groups of acoustic parameters,it suggests that the parameters of duration and fundamental frequency only influences the components of ERP in the early stage and do not influence the entire emotional cognition.(3)The difference analysis on ERP cognition process between speech intelligibility and non-speech emotions.Towards the question,the experimental design is mainly targeted on speech intelligibility and non-speech emotion cognition,to study whether there is the significant difference of speech intelligibility in emotion cognition as well as whether there is the difference between speech intelligibility and non-speech emotions in cognition process.The experimental results indicated that the P200 amplitude of speech intelligibility is the highest and the P200 latency of non-speech emotions is the shortest.According to the research,it is predicted that the latency of P200 might be related to the brain‘s participation in speech understanding and processing,and its amplitude might be concerned with the understanding degree and familiarity degree towards the speech.(4)The reconstruction and denoising methods of speech signals and EEG signals.In order to realize the simplicity and effectiveness of emotional signal recognition,it has been presented to adopt an improved compressed sensing method to make the processing and analysis and to improve the anti-noise performance of the signal.Since the compressed sensing method is not suitable for strong noise signals,this dissertation combined with the advantage of spectral subtraction method to reduce noises,two spectral subtraction noise reduction methods based on compressed sensing are proposed.Firstly,towards the problem that OMP reconstruction algorithm has quick iteration speed but the sparsity could not adapt to itself,the self-adaptive inter-frame AICSSS algorithm is presented.Secondly,towards the problem that SAMP reconstruction algorithm has high accuracy but the matching error is easy to occur,the MMOP algorithm is presented.Thirdly,integrative comparison and analysis towards the above two methods were conducted.The result shows that the AICSSS algorithm and MMOP algorithm can not only conduct accurate reconstruction of signals under the blindly sparse condition,but also improve the anti-noise performance of signals,and the reconstructed performance and anti-noise performance of the MMOP algorithm are better.(5)Single-trial ERP signal classification and identification method.This dissertation proposed a compressive sensing recognition method for single-trial ERP signals.This method mainly classifies the compressed signals and uses the K-SVD method to train the super-complete redundant sparse classification dictionary.By the residual differences between the compression signal reconstructed by the dictionary and the original compression signal,the type of the signal was determined.Experimental results show that regardless of whether the input signal is a feature or original signal,and whether the signal contains noise,the compressed sensing method can be adopted to classify signals.The recognition results obtained by using the AICSSS and MMOP methods proposed in this dissertation are superior to the traditional compressive sensing method and support vector machine(SVM)method.