Automated Music Synthesis Using Recurrent Neural Network and Long Short Term Memory

Music generation is a new domain of research in its primitive stage. Music generation model in this project employed Recurrent Neural Networks (RNN) with Long Short-Term Memory (LSTM) cells to generate melodies in the MIDI format. The model was trained on ESAC dataset of symbolic music representation using a one-hot encoding scheme. The system was able to produce musically coherent melodies with some degree of variability by adjusting the temperature parameter in the sampling function. The project demonstrated the potential of using deep learning models for music generation tasks. In this project, training accuracy of 85% was obtained.