LMD-full MIDI Dataset

The LMD-full dataset is called The Lakh MIDI Dataset v0.1, which has more than 170,000 unique MIDI files, of which 45,000 files are matched to the million song dataset. The goal of this dataset is to facilitate large-scale music information retrieval, both notation-based (using only MIDI files) and audio content-based (using information extracted from MIDI files as annotations for matching audio files).