Skip to main content
Audio to MIDI extracts MIDI notes from an audio recording using AI pitch detection. The model is Gridshift’s own, trained in house on music rendered from hundreds of sampled instruments. It is polyphonic — chords, melodies and several instruments at once — and it runs locally via CoreML. No audio is uploaded.

​
Convert audio to MIDI

1

Select an audio clip

2

Open the Command Palette and search

Open the Command Palette (⌘K) and run Convert Audio to MIDI.
Gridshift creates a new instrument track with the extracted MIDI clip; the original audio clip remains unchanged. Only the part of the file the clip actually plays is analysed, so trimming a clip before converting keeps the unwanted material out of the result. A clip warped in Smooth or Pitch is converted onto the clip’s own timeline, and Pitch is transposed to match what you hear. Punchy is not: it moves each transient independently, so its notes come back at the unwarped positions and the timing is approximate.

​
What it is good at

Sustained and plucked single-instrument material transcribes well — piano, electric piano, synth lead, bass, voice. Two known limits:
  • Strummed chords. Notes that start within a few milliseconds of each other can be merged or misplaced, and the first note of a strum is the least reliable.
  • Note lengths. Starts are far more accurate than ends. Expect to correct the lengths.
Velocity is predicted, not measured from the peak — the model was trained on how hard each note was played, so it tracks dynamics rather than the level of the recording.
Last modified on September 6, 2026