AcoustID matches known recordings, not AI-made ones

The system

AcoustID is a sister project to MusicBrainz, the open music metadata database, providing what its own site describes as 'a complete audio identification service, based entirely on open source software.' The service has three parts, per AcoustID's own description: a client library that generates compact fingerprints from audio, a crowdsourced fingerprint database, and a web service that lets applications search that database. Many fingerprints in the database are linked to MusicBrainz identifiers, connecting acoustic identification to MusicBrainz's detailed recording, release and artist metadata. The service is operated by AcoustID OÜ, a company distinct from the nonprofit MetaBrainz Foundation that runs MusicBrainz itself.

What the documents establish

The fingerprinting is done by Chromaprint, which AcoustID's own project page calls 'the core component of the AcoustID project,' an open-source, client-side library, created by Lukáš Lalinský, that implements a custom algorithm for extracting fingerprints from raw, uncompressed audio; the library does not decode compressed formats itself and relies on the calling application to supply decoded audio. Chromaprint is distributed as a C library, a Python wrapper, and a command-line tool called fpcalc, with source available on GitHub. AcoustID's web-service documentation states the lookup service is 'provided for free for non-commercial use only,' caps requests at three per second, and requires a paid commercial license for use in commercial applications, with separate API keys required for lookup and for submitting new fingerprints.

Craft and rights

AcoustID answers one narrow question well: is this audio the same recording as one already in the database. That is useful for a producer or archivist reconciling a personal library against MusicBrainz metadata, or for a developer building a tagging tool, but it is not a tool for judging whether a piece of audio was generated by AI, since Chromaprint fingerprints known recordings by their acoustic signature rather than analyzing how the audio was produced. A track with no match in AcoustID's database could be an obscure but entirely human recording, a new legitimate release not yet submitted, or an AI-generated file, and the service's own materials do not claim to distinguish between those cases. Rights holders should treat an AcoustID match as evidence of prior identity, not as a signal about a recording's licensing status or its method of creation.

Outcomes and open questions

Because AcoustID's database is crowdsourced, its coverage of any given catalog, including AI-generated tracks uploaded to streaming services, depends on whether someone has already submitted a fingerprint for that recording, which the project's own materials do not quantify.

• Does a 'no match' result in a fingerprinting database mean a recording is new, obscure, or simply unsubmitted?
• Is the tool being used to identify a known recording, or being asked to answer a different question it was not built for?
• Does the intended use fall under the service's free non-commercial terms, or does it require a commercial license?

An open fingerprinting service is a useful piece of infrastructure precisely because its job is narrow: matching known audio, not judging how any audio came to exist.