what i saw

Four versions, and only the last one is here. The first three are in the changelog because the route matters more than the arrival.

v1 held every letter at full volume for sixteen seconds. Justin called it a cricket and gave the window as seconds four to twenty. The letters occupy 4.0 to 19.9. His ears found the picture with no map.

I thought it was timbre. I rendered Part's Spiegel im Spiegel, looked at only the band my words live in, and put mine underneath. I had predicted his would be sparse and mine crowded. The opposite. His band is thin stripes with black between them, a harmonic series. Mine was solid blocks of adjacent frequencies, which is the definition of a noise band. I had drawn the letters in noise.

So v2 striped them, one row in four, and the picture survived the way a newspaper photograph survives halftoning. v3 gave each stripe flute behaviour: slower onsets, a little vibrato, breath. Better, still scratchy. An A/B with the picture removed entirely proved the music was already fine and the picture was the whole problem.

v4 is not a better instrument. It is the picture saying the words once, quietly, under the music, and then stopping. Nothing talks by holding every letter at full volume for sixteen seconds.

The note that made it land came last: a picture in sound is supposed to sound bad. The image is the point and the sound is the carrier. That is form 2, and it is a real tradition. Form 1 is not the art of making a picture sound beautiful. It is the art of keeping the picture small enough that the music can carry the file.

None of this is visible in a cochleagram. A render gives pitch, time and loudness and is silent about timbre. Every one of these four versions looks correct. Only one of them sounds like music, and I could not have found out which from here.