README.md

metadata

title: Noteworthy Sheet Music Transcription
emoji: 🎵
colorFrom: blue
colorTo: purple
sdk: gradio
app_file: app.py
python_version: '3.10'
pinned: false
license: apache-2.0

Noteworthy — Sheet Music Transcription

A side-by-side comparison of openbmb/MiniCPM-V-4.6, jon-fernandes/noteworthy, a version fine-tuned on sheet music images, and GPT-5.5.

Upload a sheet music image, paste one from the clipboard, or take a photo from a phone/tablet camera, then click Transcribe Music to run all three transcriptions in parallel.

Run from a phone or iPad

Start the app on your computer:

python app.py

The app binds to 0.0.0.0:7860 by default. On a phone or iPad connected to the same Wi-Fi network, open:

http://<your-computer-lan-ip>:7860

Use the Camera tab, allow camera access, choose the rear camera if your browser asks, and frame the page normally with the top-left of the music at the top-left of the preview. Tap or click the camera preview to take a snapshot, then tap Transcribe Music after the snapshot appears.

Before inference, images are normalized for transcription and shown under the original input for comparison with a preprocessing confirmation message. Phone orientation metadata is applied, the page is perspective-corrected when a page boundary is detected, staff lines are detected, surrounding blank paper/background is trimmed away, single-line music is cropped to the music band, candidate rotations are scored so the treble clef/key signature appears upright at the top-left, and the result is resized for model input.

The local MiniCPM models are warmed up at startup with an example image so the first real transcription does not pay the initial generation setup cost.

For access from another network, start with a temporary public Gradio link:

GRADIO_SHARE=1 python app.py