AI music transcription that returns notes you can edit — MIDI, a real score, and a session. Each instrument goes to the model that suits it. And one line can come straight out of a full mix.
Free runs daily · browser only · nothing to install
Separating a track into stems first is still the accuracy win for a full arrangement. But it is not always what you want. If you only need the solo — the sax chorus, the guitar break, the piano line — Solo Mini reads the whole mix and returns that one instrument, with no separation step in front of it. Naming the instrument does the work the separator would have done.
It is tuned on fast solo playing, which is the material a general transcriber handles worst: dense lines where the notes arrive faster than the model wants to commit to them. On held-out material it roughly doubles onset accuracy against the same base model for piano, saxophone and trumpet.
It also fixes the characteristic failure of an untuned model pointed at a full mix — transcribing everything it can hear at once. On one Charlie Parker chorus the untuned model returned 2,232 notes where the human transcription has 389. It was not hearing the wrong pitches; it could not tell which line it had been asked for.
Saxophone, trumpet, guitar and piano. In beta, and a specialist rather than a replacement: for sustained, sparse or ensemble material the standard route is still the better one. Bass keeps the standard route — the tuning did not improve it, so it does not get it.
A stem, a solo, or the whole track. Anything the browser can read.
That single choice picks the model — and for a solo, it is also what pulls the part out of the mix.
Editable MIDI, a real score with clefs and key signatures, and a session for Logic or Ableton.
A piano take and a solo horn are not the same problem, and one general transcriber compromises for both. So the take is routed: a dedicated polyphonic model for piano, register-conditioned pitch tracking for solo winds, brass, strings and bass, a general polyphonic model for guitar and ensemble material — and Solo Mini when you want one line out of a mix. Tell it what you recorded and the routing follows; if you do not, it is inferred from the audio.
Notes without a grid are a list, not a transcription. Tempo, downbeat and meter are detected first and the notes are fitted against them — and the tempo map is allowed to move, so a performance that slows into a cadence transcribes at the tempo it was actually played instead of being flattened onto a click.
Notes, as editable multitrack MIDI you can correct by hand. Sheet music, as a multi-stave score with clefs, key signatures and written transposition for transposing instruments. MIDI per part that reflects your edits rather than the first guess. And a Logic or Ableton session with the stems on named tracks, the tempo map and a chord guide.
It converts recorded audio into symbolic music — notes, rhythms, bar lines and chords. The output is editable MIDI and notation, not a text document.
Usually it helps, and for a full arrangement it is still the biggest accuracy win. But for a single solo line, Solo Mini reads the mix directly: you name the instrument and it returns that part, with no separation step.
Saxophone, trumpet, guitar and piano. Bass keeps the standard route — the tuning did not improve it, so it does not get it.
It depends far more on the material than the model. A clean solo or a single-line horn transcribes very well; a dense, reverberant mix is genuinely harder because the notes overlap in the signal. Every result stays editable.
No. It runs in the browser. Free runs every day with a free account.