Find out what the channel means
TouchDesigner’s audioAnalysis palette component provides several audio-derived channels, including band levels and spectral measurements. Derivative recommends activating only the quantities needed to reduce computation. Begin with one measurement and inspect its output while listening to the actual source you intend to use.
Play a sparse phrase, a dense phrase and silence. Observe the channel over time. Does it return to a useful baseline? Does it mostly sit near one end of its range? Does a supposedly rhythmic measurement react to an unrelated sustained sound? These observations matter before the value is connected to a visible object.
Name the channel after its role in your patch, while preserving a note of the original measurement. A name such as pulse-energy can describe your interpretation, but it should not conceal whether the value came from a low-frequency band, an onset detector or a manual control. Keep that distinction available for later debugging.
Shape the response deliberately
First map the measured range into a restrained visual range. A signal spanning zero to one does not imply that the destination should use its full available travel. Choose a minimum that leaves a recognisable image and a maximum that preserves the composition. Test quiet material as carefully as loud material.
Then consider time. Derivative’s Lag CHOP allows different response times for rising and falling values. A fast rise with a slower fall can make an event clear without forcing the image to collapse immediately afterwards. Equal smoothing in both directions produces another character. Neither is inherently more musical; compare them with the phrase.
Watch the raw and shaped channels together. This reveals whether an apparent delay comes from the input analysis, the smoothing you added or the visual rendering. Without that comparison, it is easy to compensate in the wrong place. Keep the original measurement available rather than overwriting it with several transformations.
Use thresholds only when the visual action genuinely needs a decision. A continuous change in size can tolerate variation that would make a scene trigger chatter repeatedly. For a discrete action, design a clear re-arming condition as well as the trigger threshold. A sustained loud passage should not accidentally count as many separate entrances.
Give the measurement a visual responsibility
Connect the shaped channel to one visual property. Choose a responsibility that suits its behaviour: a slow measure might influence the field’s openness, while a brief event could mark an entrance. Avoid using a single channel to control every visible property merely because the connection is easy.
Add a manual override that can hold or reduce the response during performance. This is not an admission that the analysis failed. It gives the performer a way to interpret the music rather than surrendering the whole presentation to a measurement. Rehearse entering and leaving the override so it does not create an unexplained jump.
Test at the intended output resolution with the rest of the project running. Inspect performance during the busiest passage and while changing scenes. A successful measurement in an empty network does not establish how the finished set will behave under load.
Save a short diagnostic recording or screen capture showing the input, raw channel, shaped channel and result. Alongside it, write the musical purpose in one sentence. That small record will help when you return to the patch weeks later. It also gives collaborators a clear starting point: what is measured, how it is transformed and why the image responds as it does.
Try this.
Use one audioAnalysis output, inspect it during silence and two contrasting phrases, then compare raw and Lag-shaped versions. Map the result to one visual property and add a rehearsed manual override.
Listen back
- What does the channel actually measure?
- Where is the apparent delay introduced?
- Does this visual action need a continuous value or a discrete decision?
