Loading Musirion

Blog

November 13, 2026

Introducing Musirion transcription model 2.0

We have released Musirion transcription model 2.0. It is based on a completely redesigned architecture and produces sheet music of significantly higher quality than previous versions.

A live performance cannot simply be transferred to notation one note at a time. Musicians deviate from the metronome, hold notes for different lengths, add pauses, and vary the tempo. A literal conversion of MIDI events results in incorrect durations, unnecessary rests, and rhythms that are difficult to read.

Model 2.0 analyses the musical context of a piece instead of treating each note in isolation. It reconstructs rhythm and note durations, identifies measure boundaries, separates the musical material into voices, and distributes it between staves.

We have also completely redesigned the way the score is constructed. The new model makes connected decisions about the structure of the notation, allowing it to handle complex rhythms, chords, multiple voices, and expressive timing deviations more effectively.

As a result, scores are more accurate, cleaner, and noticeably easier to read and continue working with.

The new architecture also creates a stronger foundation for future improvements to rhythm, voices, score layout, and musical markings.

Model 2.0 is now available in Musirion. Select it when uploading a new recording and try it with your own performance.