What’s Out There
← Back to swiping
Audio & musicFreeAcademic

MusicNet

Thickstun, Harchaoui, Kakade (University of Washington)

330 classical recordings with over a million labeled notes aligned to the second.

A collection of 330 freely licensed classical music recordings with more than 1 million labels giving the precise time, instrument, and metrical position of each note. Labels come from scores aligned with dynamic time warping and verified by trained musicians, with an estimated error rate of about 4 percent. Used for note prediction and transcription research.

Access

Cost
FreeFree download from Zenodo; no account needed.
License
Open licenseCC BY 4.0
Sign-up
No sign-up
How you get it
Bulk downloadCSV, GZ (WAV and labels)

What's in it

One row is
One recording with note-level labels
Covers
Not geographic
Years
Release 2016
Updated
One-time release
Size
330 recordings, more than 1 million labels
note start timenote end timeinstrumentpitchmetrical positioncomposition

Quality

4 of 5

Influential benchmark, with a stated error rate; static since 2016.

  • Labels have about 4 percent error
  • Mostly Western classical works

Connects to

  • GuitarSet: Smaller dataset focused on guitar transcription
  • MAESTRO: Aligned piano recordings with MIDI
  • MUSDB18: Mixtures with stems for separation tasks

Questions it could answer

  • How accurately can models transcribe ensembles?
  • Which instruments are most often confused?

Similar data sets

  • MusicBrainz Database

    An open, crowd-edited encyclopedia of recorded music, with the core data in the public domain.

    Audio & musicMetaBrainz FoundationFree

  • GuitarSet

    360 guitar excerpts with hexaphonic pickups, giving note-level labels for each string.

    Audio & musicXi, Bittner, Pauwels, Ye, Bello (NYU MARL)Free

  • AudioSet

    Two million 10-second YouTube clips labeled with 527 kinds of sound, from barking to bagpipes.

    Audio & musicGoogle Research (Sound Understanding group)Free

  • FSD50K

    51,197 Freesound clips labeled with 200 sound-event classes drawn from the AudioSet Ontology.

    Audio & musicFonseca, Favory, Pons, et al. (Universitat Pompeu Fabra)Free

  • Million Song Dataset

    Audio features for a million songs, with no audio, frozen in December 2010.

    Audio & musicLabROSA (Columbia University) and The Echo NestFree

  • MUSDB18: a corpus for music separation

    150 full songs delivered as a mix plus separated vocals, drums, bass, and other stems.

    Audio & musicRafii, Liutkus, Stöter, et al. (SigSep)Free