HarmonyRFG: Bridging Music Theory and Deep Learning through Spiking Neural Networks for Innovative Chord Progression Generation

Abstract This research proposes a new music generation model Harmony Rule-based Forward Generation that integrates traditional harmony rules and deep learning. The model combines the global modeling capability of Transformer with the temporal perception characteristics of Spiking Neural Networks, effectively improving the artistry and logic of generated chord sequences. By introducing Markov chains and music theory rules, it achieves automatic generation of chord sequences that conform to human intuitive perception and possess innovation, and in the meantime, showcases the application potential in real-time composition, interactive music installations, human-computer collaboration, and other artistic scenarios.