Multi-category MIDI music generation based on LSTM Generative adversarial network

Music generation by neural networks has become a central issue since deep neural networks demonstrated their ability in learning from big data collections. This paper proposes a music score generation model which employs multi-layer RNNs and GAN scheme. First of all, the midi sequences are passed to the model, which is parsed as tone lengths, frequencies, intensities, and timing, and then the music theory law is introduced, while the initial sequences are set as music chords. Consequently, the distribution of music is learned in the process of training. The experimental results show that it is a feasible network structure which can generate multi-category music with good hearing experience.