Explore high-quality datasets for your AI and machine learning projects.
The repository collects various symbolic music datasets, including piano‑roll datasets, lead‑sheet datasets, and MIDI datasets, each with its specific source and format.
The dataset named maestro‑base‑v2 is intended for music analysis. It includes three main features: `notes`, `control_changes`, and `source`. `notes` contain numeric fields for note end time, pitch, start time, and velocity. `control_changes` contain numeric fields for control number, time, and value. `source` is a string possibly indicating the music source. The dataset is split into validation (137 samples), test (177 samples), and train (962 samples). Total download size is 141,530,448 bytes; total size is 493,963,458 bytes.