AI MUSIC RADAR
On-device music transcription

Play it. See it.

AI Music Radar turns what you hear into sheet music — live and entirely on your iPhone. For musicians who learn and write by ear: notes appear while you play, then you get a real engraved score, a piano roll, and MIDI / MusicXML / PDF export.

On your phone? Watch the 19-second vertical cut.

Why it's different

Three things no competitor in the category puts together.

01 · LIVE

Notes while you play

The web demo transcribes at about 0.7 s latency. No competitor checked on iOS does real-time polyphonic transcription.

02 · ON-DEVICE

Private by architecture

The Core ML model (Spotify Basic Pitch) runs locally. It works offline with no login — your audio never leaves the phone, and no server outage can take it down.

03 · REAL OUTPUT

Made for musicians

An engraved staff, tempo / meter / key estimation, playback with note highlighting, and MIDI / MusicXML / PDF that opens in MuseScore, Dorico, Sibelius or Logic.

Screenshots

From the App Store listing — the app as it runs on iPhone.

Hear it
Play it. See the notes. Live.Tap record — the score writes itself.
Sheet music
Runs on your iPhone.No upload, no account.
Playback
Engraved score, in key and in time.Tempo, meter and key detected.
Piano roll
Playback that follows every note.Watch each note light up as it sounds.
Export
Export MIDI, MusicXML and PDF.Take it to your DAW or notation app.

How it works

Mic to score, start to finish, on the Neural Engine.

Microphone
live audio in
22 kHz mono
capture
Core ML Basic Pitch
Neural Engine · 2-sec windows
Swift decoder
posteriors → notes
Quantizer
tempo · meter · key · 16th grid
Engraved score
staff + MIDI
No audio ever leaves your phoneWorks in airplane mode

Under active tuning

The model is Spotify's Basic Pitch (ICASSP 2022 weights), pinned in models.lock with SHA-256 verification, shipped as versioned release assets — never in git.

Recent work tuned the decoder, not just the model — onset 0.7, frame 0.4, min note 5 frames — lifting note accuracy on real recordings with no retraining. Next: real dynamics. An audit found velocity was tracking model confidence, not loudness; a per-pitch energy calibration now beats both the old estimate and a trained neural head.

84.3
GuitarSet note F1 (was 80.2)
73.5
MAESTRO note F1 (was 69.5)
76.2
SMD note F1 (was 70.3)
9.2
velocity error, calibrated (was 19.4)

Pricing

Because transcription runs on-device, the marginal cost is roughly zero — so there's a lifetime option almost nobody in the category has.

Free
$0
forever
  • Unlimited live transcription
  • 3 saved pieces
  • 30-second PDF export
Pro
$4.99/mo
or $29.99/yr · 7-day free trial
  • Unlimited saves
  • Full-length import
  • MIDI + MusicXML + full PDF
  • Key / meter / tempo edits, slow playback
Lifetime
$49.99
one-time · launch price $39.99
  • Pro forever
  • No subscription
  • No credits that don't renew

Hum the song stuck in your head.
Get the notes in seconds.