KRAISLER: A Multi‑Track Dataset of Piano and Violin Duet Recordings for Music Information Retrieval Research

The piano plays a central role in classical music, both as a solo instrument and in ensemble settings, often accompanying melodic instruments such as the violin, cello, and flute. Recordings of piano performances by human players in these musical arrangements have been valuable resources for music information retrieval (MIR) research, supporting tasks such as automatic music transcription, performance analysis, and automatic music accompaniment. However, most existing datasets focus on solo piano performances or provide only mixed audio in ensemble contexts, limiting the usage for various scenarios of music analysis. We present KRAISLER, a new multi‑track dataset of piano and violin duets featuring high‑quality audio from simultaneous recording of live studio ensemble performances by professional musicians in acoustically isolated rooms. The dataset comprises 20 excerpts from pieces spanning a wide range of historical periods and classical genres. Each excerpt is accompanied by directly captured piano MIDI data, musical scores with corresponding score‑alignment data, note annotations, and beat annotations. We evaluate the dataset across multiple MIR tasks using state‑of‑the‑art models, including piano or violin transcription, piano–violin source separation, score alignment, and beat tracking, and establish it as a benchmark. Beyond the tasks applied in the benchmark, we expect the dataset to be well‑suited for various expressive performance and ensemble research topics.