Music transcription researchOperational R&D

CueForge Score

A performance is not the composition.

An ambitious musician-guided transcription and notation project. Its working engine generates editable scores, while its research asks how to distinguish written composition from valid performance interpretation.

Visit the real website ↗Read the story ↓

Score notation editor with a labeled sample MusicXML score

PROJECT STATELive application · accuracy below target

VISITOR EXPERIENCECase study now · separate demo later

DISCIPLINEMusic transcription research

THE STORY BEHIND THE WORK

Why we
built it.

Good tools start with something somebody was actually trying to do.

01 / WHERE IT STARTED

A recording is evidence of a performance, not a transparent copy of its written score. Translating it to useful notation raised a harder question: what part is composition, and what part belongs to the musician?

02 / WHAT GOT IN THE WAY

Pitch detection alone cannot tell us how music functions. Timing, phrasing, overlapping instruments, production, notation and performance choices can be mistaken for score facts.

03 / WHAT WE LEARNED

One celebrated recording cannot be the universal answer. Our Reference Lab framework compares three independently observed interpretations against a verified working score, preserving variation instead of averaging it away.

04 / WHAT WE BUILT

A deployed audio/stem processing path, MusicXML generation, review, notation editor, comparison and export workflow, plus a separate six-module Reference Lab with source/provenance controls and human-gated promotion.

05 / WHAT CHANGED

The processing pipeline was made operational. Benchmark tests, including Ode to Joy, demonstrated that an apparently successful export can still contain incorrect notes, duplicated parts, and truncated music.

06 / WHERE IT LED

This remains our substantial R&D case study. The product is operational, but transcription accuracy has not met the standard for release as an accurate one-click system.

UNDER THE SURFACE

More than
a landing page.

These are the core components represented in the supplied application or documented development history—not an assertion that every planned feature has passed independent production acceptance.

Audio and stem preparationSeparate processing / engraving engineMusicXML editor and exportThree-performance research framework

01 / ORIGINAL DEPLOYMENT

Visit the real website.

Live development application; processing accuracy remains experimental.

Visit the real website ↗

02 / INDEPENDENT DEMONSTRATION

Explore the workflow.

Independent, synthetic-data demonstration in preparation. No production database, private content, payment or outbound notification will be exposed.

RESEARCH METHOD / THREE INTERPRETATIONS

One composition. Three performances.

Three independent observations help us distinguish stable score evidence from meaningful timing, articulation, phrasing, dynamics and production differences. This is a research framework—not a claim that full comparative listening or accurate automatic transcription is complete.

REFERENCEVerified working score

Structure, notation, instrumentation and musical landmarks.

OBSERVEThree independent interpretations

Keep variation visible instead of averaging it into one supposedly correct performance.

REVIEWMusician-led validation

Candidate findings require review before any production promotion.

Why we paused: The processing path works, but transcription quality has not met the intended musical standard. No one-click accuracy is claimed. The Reference Lab remains private and separate from the deployed application.