AeberScale

AeberScale is a scale finding algorithm that modifies the famous Krumhansl-Schmuckler key-finding algorithm for jazz music. In particular, rather than using solely major or minor key profiles, we adapt the scales from the (equally famous!) Jamey Aebersold Scale Syllabus. This syllabus is perhaps best known to musicians for its inclusion in copies of the Charlie Parker Omnibook (check the back page!).

Usage

AeberScale is an extremely simple package that has no external dependencies. It relies on you providing it a set of pitch classes and their corresponding durations. It is up to you how you extract this information, so you might find it handy to use a package like pretty-midi or music21, depending on what sort of input you are using.

Once you have this information, you can use aeberscale.find_key to estimate the key:

from pretty_midi import PrettyMIDI
from aeberscale import find_scale

# Extract notes and durations from MIDI file
pm = PrettyMIDI("path/to/a/midi/file")
nots = [p.pitch for p in pm.instruments[0].notes]
durs = [p.duration for p in pm.instruments[0].notes]

# grab the scale with aeberscale
out = find_scale(nots, durs)
print(out)
>>> "F major"

Note that the returned object from find_scale is an instance of a subclass aeberscale.Scale. This provides some useful helper properties and methods:

print(out.notes)
>>> ["F", "G", "A", "A#", "C", "D", "E"]

print(out.note_numbers)
>>> [5, 7, 9, 10, 0, 2, 4]

print(out.family)
>>> "major"

You can also easily convert from note names or chromatic scale steps to diatonic scale steps (useful in, e.g., $n$-gram extraction and computation). Notes that are not contained within the current key will be returned as None:

ss = ["C", "D", "Eb", "F", "D", "Bb", "C"]
dia = out.notes_to_diatonic_scale_steps(ss)

print(dia)
>>> [5, 6, None, 0, 6, 4, 5]

Installation

Easy:

pip install aeberscale

To run the tests:

git clone https://github.com/HuwCheston/aeberscale.git
cd aeberscale
pytest tests

Algorithm

The process is as follows:

• For provided notes, obtain a histogram of summed pitch-class durations (e.g., note 5 held for X seconds, note 6 held for Y...)
• For all scales in the Jamey Aebersold syllabus and all possible root notes, obtain a binary distribution (key profile)
  • Where 1 == PC contained in scale, 0 == not (and PC \in {0, 1, 2, ..., 11})
• Compute the maximum possible correlation coefficient between the provided input and all scale/root note combinations

For more information on the Krumhansl-Schmuckler algorithm, check out this paper. Note that the only difference between the original implementation and AeberScale is that we use a binary distribution for the key profiles, rather than obtaining these through perceptual experiments.