(Background: I was asked recently for a writing sample, and I took the opportunity to restate, more concisely, what I’m trying to do with Keyboard Philharmonic.)

Musicians performing classical music and opera, and teachers and students of this music, are on the cusp of a transformation from printed music scores to digital scores. This will be as significant as the shift of text communication from printed books and magazines, to web articles, blogs, emails, and tweets.

I believe a particular model is the right next step. I call it, “Labour, symbols, and free”. It is a policy package for a music score transcription effort. It fills a gap in the present situation, and opens a gate to move forward. I will also describe the strategic context.

Obstacles and opportunities in the present situation

Unlike text, images, and video, the music score transformation suffers from a lousy revenue model. The primary customers — musicians, teachers, and students — have no money. The plan must allow for this.

There is an important asset: many of the most important musical works have printed scores in the public domain. There is no IP obstacle to transcribing them. Over 400,000 scores (7m pages) are easily accessible as free-libre scanned images, e.g. from IMSLP, CPDL, and many libraries.

Incumbent music score publishers offer printed score products: a mix of fully public-domain, fully copyright, and hybrid. They may see free digital scores as a threat; none seems to embrace them.

Notation formats adequate to enable digital scores exist: MusicXML and MEI, both XML-based. Both of them have flaws, but are 80% good enough now, and are improving. Many other minor formats also exist.

Notation editors are plentiful. They include proprietary, free-software, client-based, and SaaS-model. All have flaws, but they allow progress. They accept input via GUI, MIDI keyboard, text file, and OMR.

Optical Music Recognition (OMR) software converts an image of a printed score into symbolic music notation. Analogous to OCR for text, its accuracy is low enough to be unhelpful. Statistical techniques and deep learning promise better OMR, but this requires training data with ground truth digital scores.

Walled gardens of proprietary digital scores, e.g. Musicnotes, SmartMusic, NoteFlight Learn, sell mostly music education. Their need to pay for labour, and their incentives not to share work, limit corpus growth.

Labour, symbols, and free: getting through the gate

The gate to the transformation is to create a corpus of free-libre digital score content, transcriptions of (eventually all) public domain printed music and opera scores from the 16th–20th centuries. As this corpus grows, it allows musicians, teachers, and students do their music-making somewhat better. It also accelerates the creation of software tools for digital scores. These are the phases of the strategic context.

The innovation is the insight that the fastest and best path through this gate is immediate harnessing of voluntary labour to transcribe legacy public domain printed scores into corresponding digital scores, and to give them away for free. Think of it as: Project Gutenberg meets Wikipedia, and gets to work on music scores. This package fills a gap in the ecosystem, and has the promise to succeed where earlier efforts have failed.

It is important to mobilise labour, tens of thousands of people, because transcribing printed to digital scores at scale is a labour-intensive process. The first draft might be human effort, or ever better OMR. Either way, that draft needs human proofreading. Also, we need labour to mentor new volunteers, to administer the volunteer system, and to moderate disputes. Think Wikipedia scale.

The key to mobilising labour is social structures and norms. What is remarkable about Wikipedia is not its technology. It is the way it welcomes volunteers, funnelling them into larger and larger tasks as their interest and experience grow. Stack Overflow guides and motivates its community through points and badges. We need something like this.

We need editorial standards. e.g., a “Fidelio edition” is a digital score which faithfully reproduces a specific printed score edition. Fidelity is a bright-line standard which permits proofreading at scale by distributed volunteers. It gives musicians confidence that a digital score is as correct as its print ancestor.

The scores must have symbolic music notation content, what I call a “digital score”. These should be in widely-supported interchange formats, meaning MusicXML and MEI for now. The symbolic content is what is accessible to software tools, and this is what opens the gate. Neither images of scanned print pages, nor clean, re-engraved, human-readable scores in PDF form, are adequate.

The work product must be freely available, as public domain (or perhaps CC-BY). Why?

  1. To unlock volunteer labour. Participants will donate huge efforts to make a product which is useful and freely available. Wikipedia is a shining example.
  2. To be easily useable by anyone for any purpose. Restrictive licensing kills re-use. The walled garden is not the model which will transform music-making.
  3. To accept that our customers see the primary value in a digital score as the artistic quality and beauty put there by the composer, who is not collecting royalties. They are unlikely to pay for the transcription.

There are a two projects which have chosen the “labour, symbols, and free” package for transcribing scores:

  1. Keyboard Philharmonic. Founded by this author. Working behind the scenes, not yet at scale.
  2. Open Score project, by MuseScore. Announced Jan. 2017. Excellent intentions, no track record yet.

Several projects have/had only part of the package. Project Gutenberg produced symbolic scores freely licensed, but could not assemble the labour at scale. The Mutopia Project has scaled to thousands of scores, with free licences, but aims for formats which humans read and software does not. Numerous academic projects, like Freischütz Digital, or Stanford’s MuseData, have created fantastic digital scores, but don’t aspire to mobilise thousands of participants, nor to licence their scores freely for commercial purposes.

Strategic Context: Phases of transforming music-making

“Software is eating the world” (Mark Andreessen, 2011). The transformation of music-making represents software eating music scores, and the ways musicians use them. Why is this inevitable? A music score is an information product. A printed book is a 16th-century container for information products. A symbolic digital file is the 21st-century container. It is inevitable that music scores will move from the old container to the new. That change will ripple through musicians, and their marketplaces.

This, briefly, is how I see this transformation unfolding. It will be a cascade of overlapping phases.

Phase 1: Scanning. A growing corpus of pictures of printed scores, freely available. Already well underway. See “present situation” section above.

Phase 2: Formats and authoring. Digital score formats: MusicXML, MEI. Notation editors: Sibelius, Finale, MuseScore, Dorico. OMR tools. Started decades ago, very active now. See “present situation” above.

Phase 3: Transcription. Produce a growing corpus of digital scores, freely available. The “labour, symbols, and free” approach aims to accelerate this phase.

Benefits for music-making: musicians can revise, customise, and fix errors in digital scores. Teachers can adjust scores for better student learning. Output as re-engraved, clean PDF files, or on paper. Tablet computers allow musicians to carry a large library of human-readable PDF scores with little bulk or weight.
Drawbacks: not much different than paper-based workflows.

Phase 4: Software tools for digital scores. The corpus of digital scores justifies development of software tools which can add value using the symbolic notation data in the scores.

Some tools I expect to see:

  • Score reader apps: give musicians the tools to rehearse and perform from scores. Zoom to adjust note size for screen size and orientation, reflowing score. Tools for annotating performance instructions.
  • Digital music stand devices: tablet computers tuned for music rehearsal and performance. Large, highly-legible screens. Built-in score reader apps. Networking for score and annotation distribution. Accept pedals or other gadgets for page-turning. Highly reliable, won’t crash during performance.
  • Score “diff and “patch”: diff compares two digital scores, giving a concise statement of differences (“diff”). Patch applies a diff to one score, generating the second score. Similar to Unix diff(1) and patch(1).
  • Melody search engine.
  • Singability checker.
  • Consistency checker for orchestral parts.
  • etc. etc.

Phase 5: Music-making improves, digital-based. The combination of the corpus of scores from Phase 3, and the tools from Phase 4, permit musicians to make music in better ways, mediated by digital content and tools.

  • Digital librarians. Distribute digital scores, not paper, to orchestra and chorus. Apply typo corrections and performance annotations before distribution, saving rehearsal time. Automate return of scores.
  • Sharing “cuts” and annotations. Opera and chorus directors routinely cut or modify scores for performance. Using score diff, a director can make a diff file with their cuts, and send to a colleague.
  • Cottage editions. Low barrier to a musician accepting error corrections or interpretations to a score, and becoming editor for a new “edition”. Easy forking and sharing, as in free software.
  • GitHub for musicians: a social website where musicians can discover and share cuts, annotations, editions. An alternative to the present oral culture of cuts and annotations.
  • Commercial scores will piggyback on tools and workflows created for the public-domain digital scores.
  • Adjacent niche business opportunities. Provide value-added services to free content or tools. e.g. bespoke music score customization for a specific performance or group, score editing services.
  • Music score marketplace opportunity: a retail channel for music scores, targeting musicians, teachers, and students. Probably with a score reader. Analogues to ebook retail, with associated reader device.