█

AI/TLDR

KobiMusic · 2026-11-16 · notable

Dettato — an open 28M-parameter model that turns any recording into MIDI

In short

Dettato is an Apache-2.0 audio-to-MIDI transcriber from KobiMusic. Four small models with 28.1M parameters in total write one MIDI track per instrument plus drums, and score 59.2 composite F1 on nine sets against 45.1 for MuScriptor 1.3B.

KobiMusic · 2026-11-16 · 4 sources checked

Dettato output for Beethoven's Symphony No. 5 shown as separate instrument tracks of notes

Drop in a song and get a multi-track MIDI file back, from models small enough to run on a CPU.

What is it?

Dettato is an open-source audio-to-MIDI transcriber released by KobiMusic on 16 November 2026. It takes an MP3 or WAV file and writes a MIDI file with a separate track for each instrument and a drum track, plus a JSON file of the notes. Code and weights are Apache-2.0, and the KobiMusic post says the model is about 65x smaller than the 1.3B-parameter MuScriptor.

How does it work?

Four models run in a chain: a 15.3M transcriber finds the notes, a 4.9M evidence repairer corrects them using the transcriber's evidence, a 2.9M grouper assigns notes to instrument tracks, and a 5.0M drum model writes the drum hits. Weights download from Hugging Face on first use, and it uses a GPU when one is available.

Why does it matter?

Musicians and music-tool builders get multi-instrument transcription without a large model or a paid API, and the outputs can be used commercially. The README is clear about the limits: pitched notes cover only the 88 piano keys, instrument naming is rough on jazz, and drums are skipped for audio recorded below 40 kHz.

Who is it for?

musicians and developers of music tools

Try it

pip install git+https://github.com/kobimusic/dettato && dettato song.mp3 -o out/

Sources · 4 outlets

Tags

  • dettato
  • kobimusic
  • audio-to-midi
  • music-transcription
  • midi
  • music-ai
  • audio
  • open-weights
  • apache-2-0
  • show-hn

← All releases · Learn AI