How to use it
- Choose a recording of a single instrument or voice (up to 10 minutes).
- Select Detect notes and wait while the audio is analysed in your browser.
- Adjust the thresholds and shortest note while watching the piano roll.
- Set the tempo if you want the notes to line up with bars in your DAW, then download the MIDI file.
How it works
The audio is decoded by your browser, mixed to mono and resampled to 22,050 Hz, the input format of Basic Pitch, an open-source (Apache 2.0) note detection model published by Spotify. The model runs in your browser with TensorFlow.js and estimates, for every 11.6 ms frame and each of the 88 piano keys, how likely it is that a note starts and that a note sounds. Those probabilities are turned into notes using the onset and frame thresholds, the shortest note length and, optionally, the melodia trick, which also picks up notes that fade in without a clear attack.
The notes are written as a single-track Standard MIDI File with an acoustic piano program. Velocity comes from the detected loudness. The tempo you set only defines the bar grid; the notes keep their timing in seconds, so they will not snap to beats unless the recording was played to a click at that tempo. Pitch bends are not written.
Results depend heavily on the recording. A clean solo instrument works well; chords are detected but sometimes with missing or extra notes; drums, distortion, reverb and full mixes produce many wrong notes. Analysis takes from a few seconds to a few minutes depending on the length and your device.
Examples
Piano recording
A 2-minute phone recording of a piano piece gives a MIDI file with the melody and most chord notes; lowering the frame threshold recovers soft notes, raising it removes ghost notes.
Hummed melody
Hum a tune into a voice memo, convert it with the melodia trick on and import the MIDI into a DAW to play it on any instrument.
Frequently asked questions
Can it separate the instruments in a song?
No. It detects pitched notes in the whole recording and puts them on one track. In a full mix, vocals, bass, chords and drum overtones all end up as notes, which is why results on complete songs are messy.
Is my audio uploaded?
No. The audio is decoded and analysed in your browser. Only the model file is downloaded from this site; your recording never leaves your device.
Why don't the notes line up with the bars in my DAW?
The MIDI keeps the real timing of the performance. Set the tempo to the song's tempo and quantise in your DAW if you need notes on the grid.
Why is there a 10-minute limit?
The analysis keeps all model output in memory. Longer recordings can exhaust the browser's memory, especially on phones. Cut long files into parts first.
Privacy
Your audio file is decoded and analysed entirely in your browser; it is not uploaded or stored. The note detection model is loaded from this site's own server.