The Patch Keeps Running After I Leave the Room
Session log — Generative Soundscape Composition, Day 4
18:42 — Back at this. Third time this week. Told myself I’d leave it alone for a few days but the 7 MHz band sounded strange tonight and I couldn’t not record it.
18:47 — Loading yesterday’s patch. Immediately remember why I closed it: the panning was spinning like a broken heading indicator. Fixed that. Clamped stereo arc to 120°. Should have done this on day one.
18:51 — Recording pre-AGC this time. Lesson learned. The automatic gain control flattens everything interesting about the signal—all the fades and swells that make propagation feel alive. Hardware bypass cable is ugly but works.
18:58 — Band is doing something weird. Faint carrier around 7.045, drifting. Probably a numbers station, probably nothing. Recording anyway. Bernie Krause would call this anthropophony even though no human is audibly speaking. The ionosphere doesn’t care about taxonomy.
19:04 — Grain settings from last session: 40-70ms, density 28/sec. Bumping density to 32. Want it cloudier.
19:07 — Problem: the RF bed sounds thin against the field recording (backyard wind, 6 minutes, captured last Tuesday when the chinook came through). Need more low-end presence in the shortwave layer. Tried pitching grains down but that just makes it muddy.
19:12 — Actually the issue is the field recording is too busy. Too much biophony—magpies kept interrupting. Switching to a cleaner geophony take. Just wind against the fence. No birds. No cars.
19:18 — Better. Much better. The RF now sits under the wind instead of competing with it.
19:23 — Added amplitude-to-chord mapping. Simple version: low signal = minor triad, high signal = sus4. Not scientific. Just sounds right. When the band fades, the harmony darkens. When a carrier punches through, the chord opens up.
19:27 — Stray thought: this is the opposite of what I did in Morse Canon Choir Loops, where the signal was the structure. Here the signal is paint, not scaffolding. The structure comes from the constraints I impose.
19:31 — Spectral centroid tracking is working. Maps brightness of the RF to voicing spread in the chord generator. When the noise floor rises, the chord gets wider. When it drops, the voicing tightens. Feels like the room is breathing.
19:34 — Let it run for 8 minutes without touching anything. Watched the spectrogram scroll. Caught myself holding my breath when the band went quiet around 19:32. The system kept going. That’s the point.
19:40 — Mistake I keep making: over-processing. Every instinct says add EQ, add compression, clean it up. But hyperreal nature sounds fake. Real environments have dead frequencies. Real static has gaps. Left it alone.
19:44 — Xenakis called this stochastic composition—probability-governed, not truly random. The difference matters. Pure randomness sounds chaotic. Weighted distributions (Gaussian curves, clamped ranges) produce what he called “controlled clouds.” My clouds are controlled. Barely.
19:48 — Exported a 12-minute render. Listened back on headphones. The seam where the RF buffer loops is audible if you know where to look, around 7:20. Nobody else would notice. I notice.
19:52 — ISO 12913-1 says the soundscape is the listener’s perception, not the acoustic environment itself. So technically this piece doesn’t exist until someone hears it. Strange to think about.
19:55 — Patch is stable. Parameters logged. Could walk away and it would keep running. That was the goal, back on day one—build something that doesn’t need me. Tonight it worked.
19:58 — Band is fading. The chord generator is responding, pulling toward minor. I should stop here.
20:01 — Didn’t stop. Added a second RF buffer from 14 MHz. Immediately regretted it. Too busy. The texture lost its centre.
20:04 — Reverted. Saved the clean version. Closed the patch.
20:06 — Session ends. Coffee cold. The backyard is quiet in a way the render isn’t.