Polyphonic AI transcription · 100% on-device

Turn any sound into
editable MIDI

Drop in an MP3 and get back real notes — pitch, timing, velocity and pitch bends. Everything runs inside your browser, so your audio never touches a server.

Tip: a clean solo instrument or vocal transcribes far better than a full mix.

track.mp3
Decoding audio…0%
track.mp3 Transcribed
0notes
range
0.0sduration
0max polyphony
0.25

Changed the sensitivity? — lower catches quiet notes, higher keeps only confident ones.

How it works

  1. 01

    Decode

    Your file is decoded and resampled to 22.05 kHz mono, entirely in the browser.

  2. 02

    Transcribe

    A neural network estimates note onsets, pitches and bends — polyphonic and instrument-agnostic.

  3. 03

    Export

    Notes are written to a standard MIDI file you can open in any DAW.

Transcription powered by Spotify’s Basic Pitch, an open-source model released under Apache 2.0.

Questions

Is my audio uploaded anywhere?

No. The model file is downloaded to your browser once, then every conversion happens locally on your machine. Your audio never leaves your device — you can even disconnect from the internet after the page loads.

What transcribes well?

Solo piano, guitar, vocals and monophonic leads give the cleanest results. Dense full mixes with drums and heavy reverb are much harder — that is a limitation of the model, not of this page.

Is it really free?

Yes. No account, no credits, no watermark, no file-size upsell. Because the work happens on your computer, serving you costs almost nothing.

What is sensitivity?

It is the confidence threshold for accepting a note. Lower values catch quiet or short notes but add false positives; higher values keep only notes the model is sure about.