Music information retrieval (MIR)
The research field that gets computers to understand music: tempo, chords, genre, similarity and what is playing right now.
The field coalesced around the ISMIR conference, first held in 2000, and the MIREX evaluation campaign, which since 2005 has pitted algorithms for beat tracking, chord recognition and melody extraction against shared test sets. The toolkit is signal processing — spectrograms, chromagrams, MFCCs — and, increasingly, machine learning. The applications live in products most people use: audio fingerprinting in Shazam, streaming recommendations, automatic beatmatching in DJ software and stem separation.
The field's chronic headache is copyright. Researchers can rarely share the audio they test on, so large datasets such as the 2011 Million Song Dataset shipped precomputed features instead of music — a million songs nobody could listen to.