Research On Music Composition Neural Network And Emotion Recognition

Music is an important form of art.With the improvement of people's daily life,the way of entertainment tends to diversify.Numerous games,animations,short videos,etc.all require a large number of original compositions to support,but the professional music production cost is higher,the style of the music is single,can not meet the individual needs of people.With the development of machine learning,computer automatic composition will greatly enhance the creative power of music,reduce the threshold of non-professional creation,and at the same time assist composers to develop new creative ideas.In order to facilitate retrieval and management,We classify the generated music emotionally.The paper analyzes the commonly used algorithmic composition methods,including the HMM model method,the music rule method,the genetic algorithm and the popular neural network composition algorithm.In order to improve the efficiency of composition,this paper uses neural network to implement algorithm composition.Based on the current research,this paper proposes a new algorithmic composition network.We obtain music sequence through an actor LSTM(Long Short Term Memory),then fix the probability of sequence by a reward based procedure that serves as feedback to improve the performance of music composition.The music theoretical rule is introduced to constrain the style of generated music.The database in the text selects classical piano music,and models the melody rules,harmony rules,arrangement rules and syncopations of classical style piano music.This paper proposes an objective evaluation method based on minimum distance,which is objectively extracted by extracting the characteristics of classical music: range,repeating notes,vertical four degrees,rhythmic variability,parallel motion,vertical three-tone,chord duration,and pitch.At the same time,the subjective evaluation system based on expert scoring is constructed.The subjective evaluation is carried out from the five perspectives of music melody,rhythm,chord,generation structure and music quality,the superiority of the algorithm is verified.The production of music depends on the burst of creators' emotions.In this paper,a differential evolution-weighted random forest classification model is proposed.Complete the emotional classification of music by integrating the time domain features,frequency domain features,auditory spectral features and nonlinear Hurst parameters of music.