Basic Pitch without TensorFlow

Audio to MIDI with Spotify's Basic Pitch. The model and weights are Spotify's, unchanged. They run here on a small C runtime compiled to WebAssembly, using several threads, with a faster note finder. Everything happens in your browser.

Audio

Drop Audio Here - or - Click to Upload

Note segmentation

How easily a note should be split into two.

0.050.95

Split notesMerge notes

Model confidence threshold

The model confidence required to create a note.

0.050.95

More notesFewer notes

Minimum pitch (Hz)

The minimum allowed pitch, in Hz. 0 for no limit.

02000

Lower notesHigher notes

Maximum pitch (Hz)

The maximum allowed pitch, in Hz.

403000

Lower notesHigher notes

Minimum note length (ms)

The minimum length required to emit a note, in milliseconds.

12500

Short notesLong notes

Also start notes where a key suddenly gets louder, not only where the model hears a note start.

Boost onsets

MIDI file tempo (BPM)

The tempo used to export the MIDI file. The notes keep their timing, only the file's beat grid changes.

24224

Notes