Music Information Retrieval
Also known as MIR
Overview
Music Information Retrieval studies how computers can identify and organize music, extract musical features, and connect recordings, scores, and descriptions to support search, analysis, and creative work. Approaches range from audio fingerprints, signal-processing features, and dynamic time warping to CNNs, transformers, self-supervised representations, and audio-language models. Common tasks include transcription and beat tracking, chord and genre recognition, music similarity and version retrieval, source separation, and recommendation, evaluated on datasets such as MAESTRO and MUSDB18. The field has strong systems for many benchmark tasks, but performance can falter on noisy or unfamiliar music and underrepresented traditions; aligning model scores with human judgments and verifying fine-grained audio understanding remain open challenges.
Latest papers
153 papers25 papers in the last four weeks, up 32% on the four weeks before. 0.1% of all new papers.