Spectral Time: hold a texture and make a clear exit

Capture something that is already sounding

This original eight-second synthetic pad contains a sustained chord with a gradual attack. Freeze deliberately starts off. Press Play, wait about one second for the chord and analysis to establish themselves, then click Freeze. Starting frozen before input has arrived can hold an empty spectrum. If that happens, unfreeze while audio is running and capture again.

Listen for a steadier spectral texture replacing some of the source's evolving detail. This is not a tiny audio loop whose waveform repeats exactly. The freezer retains analysed magnitudes and synthesizes them with advancing phases. Dry/Wet at 60% leaves some live material alongside the processed layer, so hearing the source continue does not mean Freeze failed.

While frozen, increase Rough from zero to about 20%, then return it to zero. Rough introduces phase instability into the frozen spectrum; listen for a less steady surface, not a new delay rhythm. If the hold already fills its intended space, extra instability may only distract. Reduce Rough rather than trying to disguise it with a longer release.

Rehearse the exit before extending the hold

Unfreeze around the middle of the pad. Fade In and Fade Out set how gradually the processing moves toward and away from the frozen contribution. Their panel percentages are not seconds. Start with Fade Out at 15%, compare 5% and 30%, and listen for the return of live detail. A slower change can feel gentler but occupy too much of the following phrase.

These controls smooth a transition; they do not schedule an automatic release. On unfreezing, the analyser resumes updating the captured magnitudes, so Fade Out should not be treated as a separate sampler envelope preserving one untouched old recording. Practise the move during playback rather than relying on the clip ending to demonstrate your intended musical exit.

Now leave room for the next event

The second original eight-second example is a formant-shaped synthetic tone sequence, not a human voice. Each one-second event has a gap, making overlap easier to judge. Leave Freeze off for one pass and listen to the 0.40-second spectral delay. On another pass, freeze during a sounding event, then release before the next one.

If the next event seems veiled, reduce Feedback from 20% to 5% and compare. Releasing Freeze does not empty the downstream delay memory. Feedback extends recirculation, while Dry/Wet determines how much that layer competes with the direct sequence. Reducing either can restore separation, but only the feedback change shortens the recirculating contribution itself.

Tilt is a slope in delay time across frequency, not an EQ tilt. Positive values give higher bins relatively longer delays; negative values favour longer low-bin delays. Compare +0.3 and -0.3, keeping Delay fixed, and listen to how the components spread across the gap. Spray adds differing delay offsets between bins. Back it down when you want a more cohesive echo rather than a dispersed texture.

Choose the analysis size before the performance

FFT at 2048 is the starting point, not a quality verdict. A larger size provides finer frequency spacing but greater latency and less precise timing. Changing it rebuilds the processor and clears both freeze and delay state. Make that comparison before capturing a hold you want to keep, not halfway through a transition.

Delay is quantized to analysis frames, and even its zero setting retains a minimum frame delay. The dry path is latency-aligned rather than instant. Gain adjusts the processed contribution; keep the supplied reduction or lower it if a sustained hold builds up. Clear timing and a controlled level are better stopping criteria than the longest possible tail.