| As the emotional communication medium of music producer and audience,music bears the mission of emotional expression and transmission.Therefore,this topic with deep learning method for music emotion transfer relationship is "almost the same and within the scope of the controllable fluctuations" idea into the study,in order to verify the correctness of the idea,focuses on the relationship between the expression of emotion and awakens the type of emotion,and the two kinds of emotion is defined as a two stage emotion,and provide guidance for automatic composition.The main research work is as follows:(1)A bimodal emotion fusion method with partial weight is designed to analyze expression emotion.The dual-mode emotion fusion method is composed of three parts.The first part is emotion analysis based on audio.In this part,short-time Fourier transform is used to transform audio into Mel spectrum graph,and then convolutional neural network is used to perform emotion multi-classification.The second part is the sentiment analysis based on the lyric text information.In this part,Word2 Vec is used to train the corpus,and then LSTM is used to classify the sentiment.The third part is the fusion method of two-modal emotion results.The experiment proves t hat the fusion of audio emotion and lyrics emotion can effectively replace the expression emotion of music.(2)Evocative emotion is analyzed and studied by combining Doc2 Vec and LSTM sentiment classification methods.The evocative emotion of music is obtained based on the emotion analysis of music review text.This method uses Doc2 Vec to train the corpus to obtain sentence vector,and then LSTM is used to carry out multi-emotion classification.The experimental results show that Doc2 Vec has higher accuracy than Word2 Vec.(3)Through the improved emotional representation model,this paper analyzes the emotional transfer relationship of music and discusses the inspiration and guidance of the emotional transfer of music to automatic composition.General emotio n expression methods include Plutchik three-dimensional emotion and Thayer two-dimensional emotion.On this basis,this topic introduces two indicators of positivity and intensity,improves the traditional emotion expression model,and designs emotion norm alization algorithm and emotion offset coordinate algorithm,and introduces musical emotion determination coordinate and musical emotion offset coordinate.Through the difference of two coordinates to measure the relationship between the emotional transfer ence of music.The improved musical emotion representation model can effectively reduce the number of output layers of the multi-classification model of emotion,and reduce the time and space complexity of the model.By comparing the results of musical exp ression emotion model and musical evocative emotion model,the idea that musical emotion transfer relationship is basically consistent and fluctuates in controllable range is verified,and a new idea for automatic composition is provided. |
