deepjazz

Why we included this project

deepjazz is a compact demonstration of what a recurrent neural network does with music: it trains a two-layer LSTM on a MIDI file and then generates new jazz in the same style. The whole flow, from note preprocessing to generated output, lives in a small Python codebase, so it is easy to read end to end. It dates from before large foundation models and shows the older sequence-generation approach, where a small network learns the grammar of a melody. Developers exploring generative music or sequence models will find it a useful reference, though it depends on the now-deprecated Theano backend and is not actively maintained. Its value today is mainly educational and historical, not as something to build production software on.