Adam Michael Wood
Sep 7, 2018 · 2 min read

But lots of music starts with a melody and then harmonizes it.

Most church music (hymns and chant), most choral music, a great deal of theatre and opera music and other styles of art song, and (I would guess) much symphonic music.

Even a lot of pop and folk music is melody driven. It is just that the song writers have certain chord progressions so ingrained that the same handful of progressions fit over and over.

I am a composer of mostly church music [and theatre music a long time ago]. I have always been melody driven, working out the chord progressions after the tune.

And (though it’s been a long time since I was attending composition seminars or reading advice from composers) I recall that most “great” composers focus on the melodic lines and counterpoint, rather than on moment-to-moment chord changes (though that advice probably doesn’t apply strictly to pop music).

Additionally, a song *IS* the melody, really. If you take a song and reharmonize it (for example, as Post Modern Jukebox does) it is still recognizably the same song.

But of course, a lot of tunes are obviously written over a set of changes or a riff. Especially in particular sub genres: 12 Bar Blues has a very specific sequence of chords.

But then, as you know, those chords can be substituted for chords of similar function. A good arranger doesn’t think “D — G — A — D” or even “I — IV — V — I” but something more like “Tonic — Subdominant — Dominant — Tonic”, with each of those functions encompassing a multitude of possible chords which are either consistent with those functions or which play against them.

I’m convinced all of this can be modeled in an intelligent system, but it absolutely requires a hybrid model that can move quickly between several different compositional modalities (even within the same “step”) — free creativity, constrained creativity, rote calculation, judgment of quality, etc.

I am, in a very small way, attempting to work on a piece of this problem with a (currently unpublished, but will be open source) music theory library that allows number crunching to have a better sense of the “meaning” of notes within a musical context. (I am annoyed at MIDI driven approaches which treat, for example, D-Sharp and E-Flat as essentially the same note while Middle C and an octave above middle C are essentially different notes. I think a mathematical representation of musical content which carries with it all the appropriate theoretical context is both possible and necessary.)

So I am very interested in your future work. This conversation is prodding me to at least get a beta release of my project out.

    Adam Michael Wood

    Written by