Catalogue and data licences
Where the catalogue comes from, what each dataset's licence allows, and the non-commercial basis on which Ky runs.
Updated 25 October 2026
Ky compares your file against a catalogue built from publicly available MIDI transcriptions released for research. Ky is a free, non-commercial research beta: nobody pays, nothing is sold, and no advertising is shown. On that basis it uses the datasets below under their own licences, with the attribution they ask for. We do not redistribute any dataset, and no file from the catalogue is ever served to users: a result shows an excerpt of the aligned notes (up to a few dozen bars) beside yours as an exhibit, with the transcription's title and, where known, artist and year, and the dataset and licence it comes from. Dataset licences cover the transcriptions; the songs themselves belong to their writers and publishers.
| Dataset | Licence | In the catalogue | Notes |
|---|---|---|---|
| Lakh MIDI Dataset v0.1 (lmd_full) — Colin Raffel | CC BY 4.0 | yes | ~176,000 transcriptions. Raffel, "Learning-Based Methods for Comparing Sequences, with Applications to Audio-to-MIDI Alignment and Matching", 2016. |
| Los Angeles MIDI Dataset v4 — projectlosangeles | CC BY-NC-SA 4.0 | yes (non-commercial use) | ~400,000 files; those already in Lakh are not counted twice. |
| GigaMIDI v2 — Metacreation Lab (SFU) | CC BY-NC 4.0 (gated) | yes, when available (non-commercial use) | Non-drum files not already present; Lee et al., 2025. Titles and artists from its metadata where given. |
| POP909 | MIT | optional | 909 pop songs; Wang et al., ISMIR 2020. |
| Nottingham folk collection (jukedeck/nottingham-dataset) | as published by its maintainers | test only | 1,034 folk tunes used in development. |
| MCIC (music copyright infringement cases, Park et al., ISMIR 2024) | research use, per its authors | no | Used to test the engine during development. Nothing derived from it appears in results. |
| Billboard Hot 100 chart listings | chart data compiled from public listings | n/a | Titles and chart positions used in research; no music, and nothing shown in results. |
Every result states which catalogue it was compared against and its licences, and each exhibit names the dataset its transcription comes from. GigaMIDI is used subject to its gated-access terms.
If Ky ever charges
The non-commercial datasets (LAMD, GigaMIDI) will be removed from the served catalogue, or
licensed from their owners, before any paid or advertising-supported use begins. The engine has
a switch (ky-engine run --commercial) that refuses to start on a catalogue containing
non-commercial material, so this is checked before anything is charged.
Credits and concerns
Datasets are credited here in accordance with their licences. If you own a dataset listed here and have a concern about this use, write to hello@ky.ax and we will act on it promptly.