In order to efficiently learn and generate musical nuances and complex structures that cannot be adequately represented by MIDI, data formats that extend MIDI and are more suitable for use in symbolic music generation systems have been proposed such as the REMI (REvamped MIDI-derived events) data format.REMI adds information such as note position relative to each bar, tempo, and chords, thus allowing the model to learn more musical structure during training. AI models that take REMI data as input can therefore generate music with clear harmonic and rhythmic structure, and by building a model that takes into account musical elements such as bars, beats, and sub-beats, it is possible to generate music with a consistent song structure that is typical in pop music.
Comments are closed, but trackbacks and pingbacks are open.