The Infrastructure of Notation: Songscription, Spirit Music Group and the End of the Documentation Gap

The music industry probably owns millions of compositions it would be perfectly incapable of putting on a music stand. On September 4, 2026, Songscription revealed that it had entered into licensing agreements with several independent publishers, including its first publicly announced partnership with Spirit Music Group, to use artificial intelligence to transform recordings into sheet music, tablature and MIDI files. According to Songscription CEO Andrew Carlins, well over 90% of songs written today have never been transcribed and, although that figure remains an estimate from the company’s own CEO rather than an independently established statistic, it highlights a fairly remarkable paradox since the industry has spent decades documenting with extraordinary precision who owns a composition without necessarily documenting the composition itself. For The Sync Pipeline, this is precisely what makes the Spirit agreement much more interesting than another AI-assisted transcription tool since Songscription is not trying to generate another song from the catalog, it is trying to transform what the recording already contains into a documentary layer that did not necessarily exist before and, for the first time publicly, a publisher is allowing its repertoire to enter a system specifically designed to produce this new representation of the work.

Klem Loden

9/9/20265 min read

When the composition exists, but its instructions do not

A composition can be recorded, distributed, streamed, licensed and administered for decades without there necessarily being a score somewhere capable of precisely representing its notes, chords or structure since, contrary to what the enormous amount of sheet music available on the market might suggest, a considerable portion of popular music simply never began its life on a staff.

It began in a studio, in front of a microphone, around a guitar or behind a keyboard and, as long as the recording was enough to enable its exploitation, the absence of a written representation was not necessarily a problem anyone needed to solve.

This is precisely the relationship between necessity and documentation that Songscription is beginning to change since its system transforms audio directly into sheet music, tablature, MIDI and MusicXML and, if Andrew Carlins’ estimate is even approximately representative of reality, the potential market does not lie solely in creating a better transcription tool, it lies in that enormous amount of music that is perfectly exploitable in recorded form while part of its musical information remains locked inside the audio.

The distinction can seem fairly small until we look at what the change in format actually does. A chord progression that could be heard becomes a chord progression that can be read, notes contained in the recording become MIDI, a performance that previously had to be transcribed manually becomes data that can be modified in another environment and, ultimately, Songscription creates nothing that the composition did not already contain, it simply gives another existence to information that, until now, could only be recovered by going back and listening to it.

For The Sync Pipeline, the interest lies precisely in that translation since the music industry talks extensively about how it stores, protects and administers its assets, much less about everything it never bothered to turn into an asset because the information seemed sufficiently accessible inside the recording itself. Yet, hearing information and being able to extract, modify, teach or pass it to another machine are two fairly different things and, once AI begins crossing that distance automatically, what once looked like an absence of documentation begins to look like an inventory nobody had really thought to count.

Consent before scale

The second interesting element of the Spirit agreement lies less in what Songscription can do than in the way the company has chosen to build around it.

When the service appeared in 2025, its ability to transform practically any recording into notation already raised a fairly predictable question since, technically, being able to transcribe a work and having the right to exploit what that transcription produces remain two different conversations. A few months later, while a considerable part of the music industry continues debating what an AI company can ingest, learn from or generate without authorization, Songscription has chosen an almost opposite trajectory by securing licenses before building its expansion around ambiguity.

Spirit Music Group is the first of these agreements to be made public, but Songscription says it has already entered into agreements with other independent publishers and is preparing more. Carlins also says publishers have been particularly receptive to this approach and, behind what could easily be reduced to a matter of good practice, there is actually a much more interesting structural difference since the publisher no longer enters the conversation only when it becomes necessary to prevent a technology from touching its catalog, it enters the system that determines how that catalog can be used by it.

For The Sync Pipeline, this is probably where the word consent becomes considerably more industrial than ethical. A license obviously does not make a transcription more musically accurate simply because it exists and Songscription itself acknowledges that the accuracy of its system varies as recordings become more complex, it does, however, clarify the ground on which that transcription can be produced and exploited and, in an economy where part of the debate around AI still consists of determining whether someone should have asked permission before or after building the machine, Songscription seems to have decided that the order of operations might, itself, be part of the infrastructure.

From transcription to a system capable of reading music

This is probably where the subject begins to extend beyond Songscription itself.

Play Anything, the company that now develops Songscription and Anything Piano, describes its technology as a set of transcription models capable of transforming a recording into sheet music, MIDI, MusicXML and interactive learning tools. Anything Piano then uses the structured data extracted from the audio to build an interactive piano roll, allow users to slow down or repeat specific sections and listen to their playing in order to provide real-time feedback.

Taken individually, each of these uses can easily look like another feature in a product designed for musicians, only, when we look at them from the other end of the pipeline, something begins to change since the same composition can start moving from audio to notation, from notation to structured data and then from that data into different learning or production environments without a human necessarily having to reconstruct the information manually at every stage.

In other words, the most interesting technological asset may be neither the score nor the tutorial that appears at the end, but the intermediate layer capable of understanding enough of what is inside the recording to translate it into several different languages afterward.

For a publisher, the distinction deserves a closer look since its catalog suddenly contains no new compositions and the rights it owns have not changed either, only, the number of forms in which those same compositions can become exploitable is beginning to expand. A song that previously existed documentarily only through its recording can now produce sheet music, MIDI, MusicXML or feed a learning tool and, when that transformation can be applied not to one work but potentially to thousands, what looked like a feature designed for musicians begins quietly to resemble an infrastructure layer built for catalogs.

Documenting what was already owned

For a long time, the absence of sheet music did absolutely nothing to prevent a song from existing economically. It could be recorded, broadcast, streamed, placed in a film or advertisement and generate royalties for decades without anyone ever needing to produce a complete transcription of what was contained in the recording, so presenting notation today as a condition of value or Sync-Readiness would amount to assigning it a role the market has never actually given it.

What Songscription changes is considerably more subtle since the problem may not have been that this documentation was missing, but that producing it often cost more than its absence actually prevented anyone from doing.

AI changes precisely that equation. If transcribing a work that previously required time, specific musical expertise and human intervention can now be done in a matter of minutes and then converted into several formats, part of what catalogs never documented because it was too expensive, too slow or simply unnecessary to document suddenly becomes accessible. Songscription itself says its models can now produce these transcriptions within minutes, while acknowledging that accuracy varies depending on the complexity of the recording, which puts the technology exactly where it actually stands: not in front of a machine that has finally learned to write music perfectly, but in front of a machine that is beginning to make it economically possible to document volumes of music that a human probably would never have had the time to document.

For The Sync Pipeline, this is probably where the agreement between Songscription and Spirit Music Group deserves more attention than the simple arrival of another AI tool in music publishing since the industry is not discovering a new catalog, it is gradually discovering that part of the catalog it already owns has never existed in all the forms it could now exploit.

For decades, that prevented absolutely nothing from working because the recording was enough, only, now that machines are beginning to know how to read what remained locked inside the audio, the absence of documentation no longer looks quite like something that is missing... it is beginning to look like something waiting to be extracted.

Verified Sources and References

© 2026 Klem Loden - All rights reserved.
All content, frameworks and materials are protected under U.S. copyright law.
Sync Publishing LLC | Wyoming, USA
Registered Office: Sheridan, Wyoming