Research On Piano Automatic Composition Based On Demonstration Audio Deep Learning

In the field of composing,people need to master basic music theory,music form,harmony and other professional knowledge.For ordinary users,composing professionalism and threshold are too high.Automatic composing can enable more ordinary users to participate in music production and improve the entertainment of music.At the same time,automatic composing is random,which can bring creative inspiration to professionals.Driven by new theories,new technologies and the needs of social development,AI has accelerated its development,presenting new features such as deep learning and cross-border integration.This paper combines music with deep learning,extracts note features from demonstration audio,and constructs a neural network model to complete automatic composing.The main work and innovation of this paper are as follows:(1)In the feature extraction part,referring to the MFCC design process and combing the characteristics of piano music signal,a filter array based on twelvetone equal temperament is proposed to extract the note features of demonstration audio.At the same time,alice.XPT and the corresponding alice.wav audio files are used to verify the accuracy of the proposed method.The results show that the proposed method has high accuracy in extracting the features of notes and eliminates the influence of silent segments and overtones on pitches.(2)In the part of network model construction,the cyclic neural network has memory function and is good at processing sequential data.Piano music can be seen as a sequence composed of multiple notes in accordance with the rules of music theory,and there is a certain dependence between the notes.Automated composing allows neural network models to learn these hidden rules,and then predict the generation of note sequences.On this basis,the network model designed in this paper has five layers,and the hidden layer consists of three layers of GRU.Through many experiments,the specific parameters of the nerwork model are optimized.Finally,the quality of the piano music produced reaches an ideal effect.(3)In the part of automatic composing quality evaluation,it is difficult to evaluate the quality of piano music automatically created.Develop an online audition effect scoring platform,invite music enthusiasts to score according to their subjective sense of listening.Offline performance evaluation invites professionals to designate five indicators,empowering each of them with the method of entropy weight,and then evaluating each song comprehensively.The scoring results show that the piano music created automatically in this paper has a high score,and some of the works can pass the Turing test.The combination of online audition and offline performance designed in this paper provides a new way of thinking for the evaluation of music works.