Extract

Turn recordings into progressions. Both extractors are experimental, and how well they work varies a lot between kinds of music.

From MIDI

from chord_progressions import extract_progression_from_midi

progression = extract_progression_from_midi("song.mid")

The result is a Progression with the file's tempo and named after the file. Chord durations are rounded to the closest supported duration.

Before labeling chords, the notes are simplified. These parameters control that, each as a fraction of a whole note:

Argument	Default	Meaning
shortest_note	1/64	notes shorter than this are dropped
smooth_beat	1	repeated notes closer together than this are joined into one
quantize_beat	1/2	note starts and ends are moved to the closest multiple of this

To inspect the steps, pass simplified_path to write the simplified notes as a MIDI file, and harman_labels_path to write the chord labels as a CSV file.

The file needs a tempo; when it changes tempo, the first one is used throughout.

Method: chords are labeled with the HarmAn algorithm from Pardo and Birmingham, 2002 (PDF).

From audio

Requires Essentia with its Python bindings, which is not installed with this library.

from chord_progressions.extract.audio import extract_progression_from_audio

progression = extract_progression_from_audio("song.wav")

To inspect the steps, pass plot_dir to write a plot of the pitch-class profile over time, and marked_audio_dir to write a copy of the audio with the detected beats marked. Both directories must exist.

Method: pitch-class profiles (HPCP) from Gómez, 2006, as implemented in Essentia, averaged over each bar.