A Hierarchical Recurrent Neural Network for Symbolic Melody Generation

In recent years, neural networks have been used to generate music pieces, especially symbolic melody. However, the long-term structure in the melody has posed great difficulty for designing a good model. In this paper, we present a hierarchical recurrent neural network for melody generation, which consists of three Long-Short-Term-Memory (LSTM) subnetworks working in a coarse-to-fine manner. Specifically, the three subnetworks generate bar profiles, beat profiles and notes in turn, and the output of the high-level subnetworks are fed into the low-level subnetworks, serving as guidance for generating the finer time-scale melody components. Two human behavior experiments demonstrate the advantage of this structure over the single-layer LSTM which attempts to learn all hidden structures in melodies. In the third human behavior experiment, subjects are asked to judge whether the generated melody is composed by human or computer. The results show that 33.69% of the generated melodies are wrongly classified as human composed.