Music Score Modeling And Generation By Recurrent Neural Network

The use of algorithms,or at least some of the rules of formatting,for music composition has been around for centuries.But,there are still many problems in terms of music score modeling and generation,such as whether the music format is correct,whether the rhythm is consistent with the label,whether the melody is appropriate,and so on.In recent years,with the rapid development of recurrent neural networks,the advantages in the field of time series modeling and generation have been widely concerned and applied.As one of the best methods in algorithm composition,recurrent neural networks are applied to the mission of music score modeling and generation,and have made great contributions that cannot be ignored.Therefore,this paper presents a recurrent neural network,ABC-RNN,for ABC format music scores modeling and generation.Based on the long short-term memory unit,the network uses the attention model,the dropout layer and the embedding layer.By using the attention model,the ABC-RNN can have a better recognition effect for different types of context elements.The addition of the dropout layer can solve the over-fitting problem of the RNN model and bring smaller learning error.Having the embedding layer,the input data can be transferred into a better vector performance and the learning cost of the model can be reduced.Finally,the ABC-RNN model proposed in this paper can get more efficient results than the state-of-art method.In the case of 10-round training of the data,the loss of the ABC-RNN can be reduced to 0.825 which is 9.041% lower than that of the state-of-art method,and the accuracy of music score generation can achieve 86% which is 59.259% higher than that of the state-of-art method.In addition,the ABC-RNN model proposed in this paper can not only effectively reduce the magnitude of error generation,reduce the impact of error occurrence on the whole music score,but also learn the structural features of the music score to maintain the rhythm of the same,melody harmony,and to achieve an excellent music score modeling and generation effects.