AI Music Therapist: A Study on Generating Specific Therapeutic Music based on Deep Generative Adversarial Network Approach

Music therapy models have been widely known in the medical field for their effectiveness in early childhood trauma, depression, and Parkinson's disease, and have been incorporated in the treatments of many medically advanced countries. The main musical tools in music therapy are listening and improvisation, but given that music composition requires the accumulation of much experience and time spent on arranging, neural networks are beginning to take effect in solving this problem. Adversarial Generative Network (GAN), Long-Short Time Memory (LSTM), and other neural networks learn data from prepared MIDI (Musical Instrument Digital Interface) files with therapeutic effects. Moreover, they generate recommendations of new therapeutic music of specific genres for users, helping them to perceive the style and structure of the new music. The experiments demonstrate that the designed GAN neural network makes the music more natural and smooth.