It is common to find progressions where the same roman numeral repeats itself:

I think in these cases, a common procedure for "good" voice leading is to prefer a change of the voicing (e.g., change of position in the three upper voices), penalizing the algorithm for proposing the exact same voicing twice.

I wanted to make sure that the policy coincides with the practice in existing music.

I looked for repeated harmonies like these in 69 annotated Bach Chorales and found 192 instances.

Out of those 192 instances:

• In 78% of them, the voicing changes
• In 22% of them, the voicing remains the same

Here are a few of the instances I found. Red means different voicing and green means same voicing.

For completeness, here is the full table. Offsets are given in quarter notes from the beginning of the score (starting from 0.0). In all instances, chord1 and chord2 are contiguous and annotated by a human expert:

Currently, the algorithm will always produce the same voicing twice.

Changing this should be a 2-line diff. PR coming soon.

Hi @ekzhang,

I just made a fork of your repo. I think it is a pretty cool starting point!

However, I just noticed that there is no LICENSE indicated.

Can I assume that it is okay for me to use the code and modify it? If so, maybe adding a LICENSE would help others knowing this as well.

Cheers.

In this example, the tritone (F and B) of a viio moves in parallel motion upwards. The parallel motion is probably allowed because the quality of the fifth is different diminished fifth != perfect fifth and it is not penalized as a parallel fifth.

However, the most important thing is that the F should preferably resolve to E instead of going up to G. This is already handled in V7 chords with the rule of Seventh should resolve downward or stay in unison, however, this does not mitigate the same problem in viio chords, because the F is the fifth of that triad, not the seventh. The code is looking for chord.seventh.

The example can be reproduced with this input:

parser.set_defaults(
        key="C",
        chord_progression="I IV6 V6 I V vi IV IV2 viio6 I V I V6 vi65 viio6 I6 ii65 V7 I",
        durations="1 1 1 1 1 1 1 1/2 1/2 1 2 1 1 1 1 1 1 1 2",
        time_signature="3/4",
)

In the same example, a melodic leap of a seventh is seen in the bass (D down a seventh to E). Some melodic motions in the same voice should be penalized. Sevenths are one of them. Augmented seconds are another one. Tritones are another one. Maybe more I can't remember now. This can be a separate issue.