muScribe: Automated tranScription of muSic

muScribe is poised to revolutionize music transcription by introducing advanced AI to transcribe audio recordings into detailed music scores. This project, carried out at the RITMO Centre for Interdisciplinary Studies in Rhythm, Time, and Motion at the University of Oslo, seeks to make music more accessible to the public.

Our primary objective is to develop a service for music archives to digitize music performance recording using state-of-the-art deep learning and our own cutting-edge research. Our hybrid approach is markedly original, merging the strengths of machine learning with symbolic AI, rooted in cognitive science and musicology.

The project is particularly oriented towards cultural institutions, music publishers, and copyright organizations. By automating transcription, we reduce costs and increase the precision and availability of music scores.

muScribe is a continuation of the MIRAGE project.

A demo of some of our new transcription technologies:

Automated transcription and visualisation using our hybrid AI technologies:

  • deep-learning-based note detector developed by Lars Monstad
  • signal processing and rule-based post-processing, for pitch and timing correction and string segregation.

Illustrated on the tune Sord?len, played on the Hardanger fiddle by Otto Furholt. A tune from the Norwegian folk music archive.

Published Oct. 21, 2024 12:00 PM - Last modified Aug. 8, 2025 10:23 PM

Contact

Head of project:

Olivier Lartillot

Participants

  • Olivier Lartillot University of Oslo
  • Lars Alfred L?berg Monstad University of Oslo
  • Karstein Gr?nnesby
  • P?l Br?telund
Detailed list of participants