Mari’s comment on yesterday’s post about the Candy colored musical keyboard was: “I typed different words and listened.”
Which was absolutely brilliant!!!
Today I tried typing different words and phrases into the applet, and listened to the melodies that emerged. Mostly it sounded more or less like a parody of Schönberg, with “melodies” that seemed rather arbitrary.
Then I spent some time searching for words that would produce a strong lilt. Some words, like “once”, had a lovely melody, and occasionally an entire phrase, like “who am I”, sound as though it could be the melody for a song with that title.
But for the most part the results just didn’t add up. Since my keyboard layout was mapping to the 12 notes of the chromatic somewhat randomly (given the quirkiness of QWERTY), this was to be expected.
But what if we were to map the keys in a way that tried to optimize for mapping words to good melodies? For example, we could map letters, such as “s” or “d”, that often occur at the ends of words to the tonic or dominant note of a scale. And common letter combinations like “th” could map to pleasant intervals, such as the major third.
Of course it’s hard to know, a priori, whether a mapping will be good. So one possibility would be to evolve our keyboard, starting with keys mapped to notes at random. Then we would have the computer “type” a typical text, while we listened for whether interesting melodies emerge.
Then, just as Karl Sims evolved virtual creatures through genetic algorithms, we could incrementally change our mapping in random ways, and always choose the alternative that sounded best.
Maybe I’ll try something like that.