URMP stands for University of Rochester Multi-Modal Musical Performance, which is a dataset for audio-visual analysis of music performances. The dataset includes 44 simple multi-instrumental musical works, which are composed of performances recorded separately for a single track. For each piece, the dataset provides a score in MIDI format, a high-quality recording of a single instrument, and a video of the synthesized work.

The dataset is compared with existing widely used music audio datasets and is expected to be useful for developing and evaluating existing music information retrieval (MIR) tasks, as well as new multimodal tasks. The dataset benchmarks two existing MIR tasks (multiplicative analysis and musical score source separation) and compares them with other existing music audio datasets.