Adding AcoustIDs to recordings with ISRCs

Is there any way to add AcousticIDs to recordings already in MB with their ISRCs?

You can submit AcoustIDs using MusicBrainz Picard, see Submitting Acoustic Fingerprints — MusicBrainz Picard v2.6.3 documentation

But I don’t know what you want to do with the ISRCs. Can you explain this in detail?

In case a standard ID already exists for a recording, it should be possible to map an ID into another.

You mean you want to manually link an existing AcoustID to the recording? As this is about audio analysis the proper way is to analyze the audio itself and submit the data.

I think manual linking AcoustID just based on metadata won’t work well, it kind of goes against the purpose of AcoustID. I also don’t see any similar recording in MusicBrainz and hence don’t which AcoustID you would use or how to select this. You’ll need the audio anyway to identify a matching AcoustID.


Is there any other site using both ISRCs and AcoustIDs to import such information?

Maybe I just misunderstand, but I don’t really get what you want to do :slight_smile: What is the actual goal you want to achieve? Do you have files that you want to identify? Do you know there are already AcoustIDs for this audio, but they are not linked to MB?

I have no files; I would like to import an AcousticID from other sources by identifying a record via its ISRC.

Import an AcousticID to where? What is that to begin with, do you mean AcoustID? Maybe try to provide an example?
I’m totally confused here :grinning:

Anyway, AcoustIDs get assigned by the AcoustID server from the fingerprint data calculated from the actual files. As pointed out above, Picard can submit the fingerprints to the server. ISRC, even if one exist for a given file or recording has no relationship with AcoustID at all.


In general if you only have the ISRC and you want to get the AcoustID for it you would need to first query MusicBrainz for recordings with this ISRC, and then query for AcoustIDs linked to those recordings.

But if the recordings in MB have no AcoustIDs linked you’ll need the audio. Without the audio you cannot identify a matching AcoustID.


In the record description

Yes, sorry: I mean AcoustID

For each record in this release I would like to include the AcoustID in addition to the existing ISRC, like in this release.

Shouldn’t two records with the same ISRC have identical AcoustIDs?

Ideally probably yes, but I honestly wouldn’t trust the labels with that. I’m very sure people here can come up with examples where the same ISRC was used for audio different enough to result in different AcoustIDs.

But anyway, in order to get a possible AcoustID for an ISRC you’d need a database of ISRCs mapped to AcoustID, and is the only such database I know of.

And to link an AcoustID to a MB recording you need to submit an audio fingerprint, and you can’t get this without the audio.


What @outsidecontext said, you’d need to submit the fingerprints to the AcoustID server. Prior that the files should already be tagged with the MusicBrainz specific tags so the server can associate the AcoustIDs with the recordings.

Not necessarily.


This is the answer to my initial question: thanks.

1 Like

ISRCs are not unique despite what ISRC Search might claim. There are many uses of the same ISRCs of different lengths reported by both CDs and Digital that turn out not to be the same recording. So, no, 2 records with the same ISRCs might not automatically equal the same recording.


But are they errors or is there something wrong with the way a recording is identified?

It’s what the label put on the CD or digital services. It’s not our job to say what’s right or wrong. If that’s what is reported by the CD or digital store, than that’s what stays on the recordings. I’m not talking about what MB says, of course an editor could have attached the wrong ISRC. That, unfortunately, happens all the time. I’m talking about ISRCs as reported by the CDs or digital stores themselves.


There are many causes. Roughly an ISRC, a MusicBrainz recording and an AcoustID have a similar definition (which is not too surprising, as AcoustID was modelled to kind of detect the differences of MB recordings). But there are many variations that cause multiple ISRCS assigned to a single MB recording, the same ISRC assigned to different MB recordings, multiple valid AcoustIDs for a MB recording. In extension an ISRC can have multiple valid AcoustID, and a AcoustID could be linked (via MB recording) to different ISRCs.

The main difference between AcoustID and ISRC is that AcoustID is a purely technically generated ID based on the audio while an ISRC is assigned. Hence the same ISRC might get assigned to slightly different edits of the same song. Some examples that come to mind:

  • Longer extra silence at the start would still be the same ISRC and usually also the same MB recording, but would produce different AcoustIDs
  • Different mastering would mean same ISRC and same MB recording, but if the differences of the fingerprints are beyond some threshold they would produce different AcoustIDs (even though AcoustID in general was modelled to allow for those small differences)
  • Heavy compression of lossy codecs could likewise cause different AcoustIDs, but same MB recording and ISRC
  • Re-releases of recordings, e.g. best-of albums or including older recordings as bonus tracks on later releases, quite frequently get new ISRCs assigned, even if they are the exact same recording. This gives you multiple ISRCs for the same recording
  • The label might not consider small edits as different enough and just reuses the same ISRC, but MB might still consider it different recordings, as might AcoustID
  • Audio with the same beginning (first 120 seconds) and length will produce the same AcoustID, even if it differs later. This e.g. sometimes happens for “clean lyrcis” versions, but there are other cases like different fade out or edited audience applause. Those cases might or might not have different MB recordings and/or different ISRCs.
  • There are also just people working for the labels. The same ISRC could be assigned to a completely different recording purely because of human error or out of lazyness.

That’s just what came to my mind, there are for sure a lot of variations to the above.


Also, it’s good to bear in mind some people are either super sloppy or don’t have a clue how to submit fingerprints correctly. AcoustID DB (and thus MB) is plagued by incorrectly matched recordings like this one used to show just until now. Sometimes these are royal pain to sort out :sob:


This is something that will get worse with versions that only have streaming sources: even if the source recording on the server is the same, the downloaded file may differ based on compressions techniques applied due to network performance or users preferences.

I think the audio compression applied by streaming services is usually of high enough quality. And AcoustID is supposed to deal with encoding differences. In general if you rip a song from CD as FLAC, then encode it to MP3 and also download it as M4A their fingerprints should be similar enough and should all end up grouped together in the same AcoustID. But it’s all a matter of similarity, use a low enough bitrate that introduces enough compression glitches and the fingerprint becomes different enough to get assigned a new AcoustID.

I think the issue with streaming services is more that no one will be able to submit fingerprints at all due to DRM and such.