Music Informatics for Radio Across the GlobE (MIRAGE) Database

MIRAGE Database v1.1.0 Overview Welcome to Music Informatics for Radio Across the GlobE (MIRAGE), a global database of metadata describing music broadcast on internet radio stations. The current release (v1.1.0) consists of metadata (e.g., artist name, track title) and musicological features (e.g., instrument list, voice type) for 1 million events captured from 10,000 internet radio stations distributed across the globe. The metadata reflect information about each event's location, station, event stream, artist, and track. Metadata include information such as the geographic locations of stations, the artist's vocal range, a list of instruments on the track, identifiers linking event records to external resources, and much more. The v1.1.0 release provides three versions of the database: Complete — all 1,000,000 captured events. CoreMatching — events for which the stream metadata reliably matched metadata from at least one of the two core open music-metadata sources, Wikidata or MusicBrainz. AllMatching — events for which the stream metadata reliably matched metadata from all four primary external music-metadata sources: Wikidata, MusicBrainz, Spotify, and Genius. The matching subsets are accompanied by corresponding entity tables for locations, stations, events, artists, and tracks. Users who wish to access, interact with, and/or export metadata from the MIRAGE-MetaCorpus may also visit the MIRAGE online dashboard at the following url: https://mirage-project.org Version 1.1.0 Version 1.1.0 extends and revises the v1.0 release in several ways, including: incorporation of additional station metadata from Radio Browser; refinement and harmonization of station and location metadata; revised event-, artist-, track-, station-, and location-level fields; revised integration of location-level population metadata; inclusion of Complete, CoreMatching, and AllMatching data releases in both CSV and Parquet formats; updated relational entity tables and identifiers; updated documentation and field dictionaries. Attribution & Citation The MIRAGE database is distributed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. Users of the MIRAGE database should cite: Sears, David R.W. “Music Informatics for Radio Across the Globe (MIRAGE) Metacorpus v1.1.0”. Zenodo, August 11, 2026. https://doi.org/10.5281/zenodo.21891580. Users accessing the MIRAGE database through the online dashboard should also cite: Ngan V.T. Nguyen, Elizabeth A.M. Acosta, Tommy Dang, and David R.W. Sears. "Exploring Internet Radio Across the Globe with the MIRAGE Online Dashboard," in Proceedings of the 25th International Society for Music Information Retrieval Conference (San Francisco, CA, 2024). Data Sources MIRAGE combines metadata from open-access, public-domain, human-annotated, and programmatically derived sources. Open-access and public sources Radio Browser (RB) -- https://www.radio-browser.info/ GeoNames (GN) -- https://www.geonames.org/ Natural Earth (NE) -- https://www.naturalearthdata.com/ WikiData (WD) -- https://www.wikidata.org MusicBrainz (MB) -- https://musicbrainz.org/ Additional MIRAGE metadata were derived from: Internet Radio Station Stream Encoder (SE) MIRAGE Team Review (MI) Commercial sources MIRAGE also uses the following commercial resources for metadata selection, matching, and attribution: Radio Garden (RG) -- https://radio.garden Spotify (SP) -- https://open.spotify.com/ Genius (GE) -- https://genius.com/ To facilitate interoperability without redistributing proprietary metadata, MIRAGE retains selected reference identifiers and canonical names from commercial resources (e.g., artist names, track titles, IDs). Data Organization MIRAGE is organized as a relational data architecture consisting of five entity types: Locations — geographic entities associated with radio stations (e.g., city, country, region, continent). Stations — radio stations and associated broadcast metadata (e.g., name, format, genre(s), language(s)). Events — individual observations of music broadcast by a station (e.g., station-local time, stream description). Artists — artists associated with broadcast events (e.g., names, identifiers, genres, instruments, voice types). Tracks — musical recordings associated with broadcast events (e.g., titles, identifiers, languages). Each entity has a persistent MIRAGE identifier (e.g., Location_MI_ID, Station_MI_ID, Event_MI_ID, Artist_MI_ID, and Track_MI_ID) that permits records to be linked across tables and across external metadata sources. The complete list of public fields and their definitions is provided in the accompanying Fields Dictionary. Data Sets Each release contains entity tables in both CSV and Parquet formats. Complete The Complete release contains all events captured by MIRAGE: MIRAGE (csv/parquet) — 1 million events events (csv/parquet) — 1 million events tracks (csv/parquet) — 414,887 tracks artists (csv/parquet) — 260,646 artists stations (csv/parquet) — 10,000 stations locations (csv/parquet) — 4,319 locations CoreMatching The CoreMatching release contains events for which the stream metadata reliably matched metadata from at least one of the database's core sources, Wikidata or MusicBrainz, using a normalized string-matching similarity threshold of ≥ 0.90 (0-1). MIRAGE_CoreMatching (csv/parquet) — 447,238 events events_CoreMatching (csv/parquet) — 447,238 events tracks_CoreMatching (csv/parquet) — 174,137 tracks artists_CoreMatching (csv/parquet) — 76,567 artists stations_CoreMatching (csv/parquet) — 9,063 stations locations_CoreMatching (csv/parquet) — 4,031 locations AllMatching The AllMatching release contains events for which the stream metadata reliably matched metadata from all four primary external music-metadata sources, Wikidata, MusicBrainz, Spotify, and Genius, using a normalized similarity threshold of ≥ 0.90. MIRAGE_AllMatching (csv/parquet) — 139,761 events events_AllMatching (csv/parquet) — 139,761 events tracks_AllMatching (csv/parquet) — 24,698 tracks artists_AllMatching (csv/parquet) — 8,750 artists stations_AllMatching (csv/parquet) — 6,604 stations locations_AllMatching (csv/parquet) — 3,238 locations Fields Dictionary The accompanying Fields Dictionary documents the public metadata fields included in the MIRAGE v1.1.0 data release. Field names follow a consistent naming convention: Entity_Source_MetadataName For example: Track_MB_MBID indicates the MusicBrainz identifier associated with a track. The public fields dictionary should be consulted for the complete list of fields, entities, definitions, field and data types, and expected values/units. Data Access and Interoperability The CSV files provide a broadly accessible format for use with common statistical, programming, and spreadsheet software. The Parquet files provide a columnar representation suitable for efficient programmatic analysis of large tables. The relational entity tables can be joined using MIRAGE persistent identifiers (e.g., Artist_MI_ID). Researchers may therefore work with the complete denormalized corpus, the individual entity tables, or the reliability-filtered CoreMatching and AllMatching subsets. Contact If you are a copyright owner for any of the metadata that appears in the MIRAGE database and would like to request removal of metadata associated with your copyrighted material, please contact the MIRAGE development team: [e-mail]