Ivory, an AI piano transcription tool we built for composers working at the keyboard

Hello everyone,

My co-founder and I have spent the past while building Ivory (https://ivory-app.com), a tool for converting piano recordings into editable notation, and I wanted to introduce it here because this community is exactly who we built it for: people who compose at the keyboard.

The problem we set out to address is a practical one. Composers who work at the piano routinely capture ideas as audio , a phone recording, a quick DAW take , and then face the slow task of transcribing their own playing by ear before they can develop the material. Manual transcription of dense voicings, fast passagework, or pedalled textures can take longer than writing the piece itself.

Ivory is built specifically for that step.

What it does

• Converts a piano recording (MP3 or similar) into notation in the browser : nothing to install.

• Uses a model trained specifically for piano, rather than a general-purpose audio-to-MIDI converter.

• Produces editable output, exportable to MusicXML (Sibelius, MuseScore, Dorico), MIDI (any DAW), and PDF.

The intended workflow: record an idea, transcribe it in Ivory, then export to your notation software or DAW and compose from a working draft rather than a blank page.

Why we focused on piano Piano is one of the harder instruments to transcribe automatically: overlapping voices, sustain pedal blurring note boundaries, fast runs, and up to ten simultaneous notes. General-purpose tools tend to struggle with exactly these cases. By narrowing the scope to piano, we handle chords, fast runs, and pedalling considerably better than the broader tools we benchmarked against.

Honest limitations

• Piano-focused , not intended for full multi-instrument or orchestral arrangements.

• The output is a first draft; expect to correct some rhythms, re-bar passages, and make voicing/enharmonic decisions yourself.

• Free tier is capped at one minute of audio per transcription (unlimited number of transcriptions) , enough to evaluate properly on real material.

We would genuinely value feedback from composers here, both on transcription quality and on how it fits your existing notation workflow. Happy to answer questions in this thread.

Edited 1 hour ago1 hr by Ivory