Paint a Tile, Hear a Tile

Brush Size Loud Brightness

Length Iterations 40 ♫ Audio → tile 🖼 Image → tile Pitch detail Time detail Levels

Paint something, then press ▶ Play.

An AuraGraph tile is a picture of sound: left to right is time, bottom to top is pitch (from 30 Hz to 11 kHz, spaced like a piano, so every octave is the same height), and brightness is loudness. This toy runs it backwards: whatever you paint, it plays.

Brushes

Tone: a pure note. Drag sideways to hold it; drag up or down to glide.

Ladder: a note plus its overtones, which is what real instruments sound like. Brightness sets how strong the upper rungs are: low is mellow (flute-ish), high is buzzy (reed-ish).

Hit: a vertical stripe, like a drum or a clap.

Noise: a soft random smear, like breath, hiss or wind.

Erase: rub it out. Undo takes back the last stroke.

Snap ♪ keeps tones on real notes; turn it off for free, slidey pitch. The faint lines are the A notes (55, 110, 220, 440 Hz…), one octave apart.

Going the other way

♫ Audio → tile turns a recording into a tile (the first 4, 8 or 15 seconds). Paint on it, then play it. ▶ Original plays the real recording for comparison.

🖼 Image → tile reads any picture as a tile: brighter is louder, top is higher. Crop a real AuraGraph tile to just its main panel first, without the axes. Or try a photo, a drawing or your handwriting and hear what it sounds like.

Lo-fi: Pitch detail, Time detail, Levels

These change only what you hear, not the picture. They pool the tile into coarser blocks before it's played, the same experiment as AuraGraph's INVERSION-1:

Pitch detail: fewer bands, and the tune melts into a rhythmic wash. Around 24 bands the melody is mostly gone, but the beat survives.

Time detail: fewer steps, and the rhythm smears into slow chords while the tune stays.

Levels: fewer loudness steps. It matters surprisingly little; even 8 levels keeps most of the music.

How the sound is made

A picture only says how loud each pitch is at each moment. It doesn't hold the phase, the exact timing of every wave. So the toy guesses it with Griffin-Lim (1984): start with random phase, then go back and forth between sound and picture, keeping the picture's loudness each time, until the two agree. More iterations is cleaner. The result always sounds a little ghostly, like a cover of what you painted rather than a recording of it.

Everything runs on this device. Nothing is uploaded.

AuraGraph · ConsciousNode SoftWorks · idea Kham, built by Kehai Interim. See research/INVERSION-1.md for what this showed about real tiles.