To start with, there is nothing new with this undertaking. The most obvious way to assess any automated process is to give it a task where both the procedure and expected outcome are well known — a task in which humans excel already.
And composing in the style of Debussy is precisely that to any music making machinery. The French composer was the first to provide a real alternative to the German/Wagnerian Romanticism. His oeuvre has become such a distinctive landmark of the twentieth century that virtually every concert-goer will spot a Debussy piece on the radio in a matter of seconds. Not only has he managed to consistently draw audiences around his novel music, but the way he did was really enthralling — which is what brings us to the very technical and down-to-earth question: why particularly Debussy?
The very honest answer is: because it was easier. To recap:
- Debussy’s music is both hugely popular and highly distinctive. Any classical lover instantly knows upon hearing some music, not only if Debussy wrote it, but if Debussy could have been writing it. This makes it uncomplicated for any educated music listener to validate, after having watched the video, whether this experiment passed or failed;
- his style is built around an inventory of highly distinguished features: whole-tone and pentatonic scales, fragmentary melodies, subtle polyphony, parallel chords and so on; these are all, fairly clearly defined, so the chances to approximate them via parameters of a MAIDENS generator are rather high;
- to be deceivingly closer to pass than to fail the experiment — given that Debussy’s orchestration preferences are so typical and well known: we added a flute, a harp and a viola to the score. Do we need to do anything else? Not that much, actually (yes, it’s called cheating). Kind of.
So, without further ado, here is the experiment itself. The video has sound (of course) and English subtitles you can turn on.
It is debatable whether the result was a complete fiasco. By no means was it an indisputable success, that’s for sure: Debussy would have used much more harmony in the harp part, the viola is extremely unconvincing and the leading melody — while possibly Debussyesque (by scale, maybe rhythm and to a lesser degree by phrasing) — lacks the pensive tone most of us expect from Monsieur Croche‘s music. The lack of any automated control for the resulting harmony also shows: the eight bars sound a little too “imponderable” for Debussy (while he did decommission functional harmony, that was neither fully, nor continually).
However, the very fact that we were able to select — among the not so many generated variants — three melodies that we could superimpose, and they did not sound bad, nor totally meaningless, is a victory per se. Especially given that MAIDENS is not a tool meant for unattended composition. It is a computer aided algorithmic composition tool, and in normal usage the human composer would have intervened for the better (musical) good.