Technology

Piano transcription technology

Musirion is built around in-house music AI models designed for solo piano transcription. The system uses a two-stage pipeline: first it converts audio into MIDI-like musical events, then it converts those events into editable sheet music.

Two-stage transcription pipeline

Input

Audio recording

solo piano audio · audio or video file

Stage 1

Audio → MIDI

note pitches · note onsets · note durations · sustain pedal · MIDI velocity

Stage 2

MIDI → Sheet Music

measure structure · rhythmic notation · voices · hand and staff assignment · tempo · MusicXML generation

How it works

In-house music AI models: Musirion uses in-house models and a custom transcription pipeline built specifically around solo piano recordings, MIDI processing and notation output.
Stage 1: Audio to MIDI: note pitches, note onsets, note durations, sustain pedal and MIDI velocity.
Stage 2: MIDI to sheet music: measure structure, rhythmic notation, score durations, voices, hand and staff assignment, tempo and MusicXML generation.
Editable MusicXML output: MIDI can be opened in DAWs and piano-roll editors. MusicXML can be opened in notation software such as MuseScore, Sibelius and Dorico.
Current limitations: complex passages, ornaments, noisy recordings, heavy reverb and non-piano instruments can reduce transcription quality.

Technology FAQ

Does Musirion use its own AI models?

Yes. Musirion uses in-house music AI models and a custom pipeline for solo piano transcription, MIDI processing and MusicXML generation.

How does Musirion convert audio to sheet music?

The system first extracts MIDI-like musical events from the audio, then converts those events into readable notation and MusicXML.

What is the difference between MIDI and MusicXML?

MIDI describes musical events such as pitch, timing and velocity. MusicXML describes editable notation that can be opened in score editors.

Why is solo piano transcription difficult?

Piano recordings often contain dense chords, overlapping notes, sustain pedal, expressive timing and two-hand notation.