Two Clocks: Mapping Slow Air Data Against Fast Sound
Sound changes in milliseconds, air in minutes. Why environmental sensor data needs a different mapping strategy than audio — and how to let both drive the same image without one drowning out the other.
The thesis device listens to two very different kinds of signal at once. Audio moves in milliseconds. CO₂, particulate matter, temperature and humidity move in minutes — sometimes hours. Feeding both into the same visual system exposes a problem that neither has on its own: they run on different clocks.
Fast signals, slow signals
Audio analysis produces a value that is never still. A kick drum, a breath, a door closing — each arrives and decays within a fraction of a second. Visual parameters driven by audio feel alive because they move at the speed of attention.
Environmental data is the opposite. A room’s CO₂ level climbs gradually as people stay in it and falls slowly once a window opens. Temperature barely moves within a session. Mapped directly to a visual parameter, these readings look static. Mapped with the same gain as audio, the noise in the sensor becomes more visible than the actual change.
Mapping by timescale, not by sensor
The approach I am working towards is to assign parameters by how fast they should change rather than by which sensor feeds them:
- Fast layer (audio): motion, flicker, particle speed — anything the eye reads as energy.
- Slow layer (air): colour temperature, density, overall brightness — the mood of the image rather than its movement.
The slow layer needs heavy smoothing. In TouchDesigner, a long Lag or Filter CHOP removes sensor jitter so that only the trend survives. The aim is for the image to drift the way the room drifts: someone walking past should not register, but an hour of people in a closed room should.
Making slow change visible
The difficulty with slow data is that change you cannot see might as well not exist. Two ideas I am testing:
- Rate instead of value. Mapping the direction of change — air getting worse or better — rather than the absolute number. A rising CO₂ trend can shift the palette even while the raw value is still in a normal range.
- Session-relative ranges. Normalising each reading against its own range over the current session, so a small but real shift fills the available colour space instead of hiding in a corner of it.
Open questions
- How slow is too slow? If a shift takes twenty minutes, does anyone notice it happened?
- Should the air ever interrupt the sound — a sudden spike in particulates breaking through the audio layer as an alert?
Next: testing the slow layer over a full working session and recording how the image changes from morning to evening.