Skip to the content.

AI Transcription of Music (Audio to Symbolic): An Assessment of the State-of-the-Art

Summary

An independent assessment of how well available open-source and commercial AI tools transcribe music from audio to symbolic notation (MIDI/XML). Starting from existing symbolic scores, the project generates audio through synthesis, transcribes it back using tools such as Basic Pitch and Klangio, and compares the transcription to the original using similarity metrics such as Mongeau–Sankoff. Beyond comparing transcription models and APIs, the pilot explores explanatory variables such as added audio effects, and sets up a reusable pipeline for systematic, recurring evaluation — an assessment currently lacking in the literature. Deliverables are a paper, open-source code, and a blog post describing the results.

Work Package

Primary WP: WP1

Duration

2026

People

Institutions