Re-Speccy

A spectral delay · Re-series · Vulpus Labs

Re-Speccy splits a sound into frequency bands and delays each band by a different amount. A SEED setting determines the pattern of delays, so the same seed always produces the same pattern.

The result is a sound that comes back in pieces. The notes of a chord can arrive one after another; the high frequencies of a drum can ring on after its body has faded; a voice can be stretched out in time, with each part of its spectrum arriving separately.

Add feedback and each band repeats at its own rate. With FREEZE, you can capture a sound in the delays and let it play indefinitely.

Re-Speccy is a complete rebuild of Speccy, with stereo input and output, more control over the delay pattern, adjustable damping, freeze, CV inputs for the main controls, and a display showing the incoming and delayed sound.

For existing Speccy users: the final section explains what's changed.

To get started: connect your sound to IN L, and connect OUT L and OUT R to your outputs. Set MIX about halfway. Each frequency band will now come back after its own delay. Turn SEED to generate a different pattern, or raise FEEDBACK to add repeats.

How it works

Re-Speccy analyses the incoming sound in overlapping chunks, each starting a quarter of a window after the previous one. It separates each chunk into frequency bands, keeps a history for each band, and plays each one back from a different point in that history. The bands are then combined into the output.

Two things are worth knowing:

• Delay times depend on WINDOW. Each band's delay is measured in windows, in quarter-window increments. The same delay setting therefore lasts longer when WINDOW is larger. Hover over TIME to see the delay in both windows and seconds.
• The delayed sound has some latency. Even the shortest delay takes time to analyse. At the default WINDOW of 256 samples, this adds 6.7 ms; at 2048 samples, it adds 53 ms. The dry signal has no added delay.

BANDS

WINDOW sets the size of the analysis window, and therefore how finely the sound is split into frequency bands. There are six settings: 64, 128, 256, 512, 1024 and 2048 samples. The default is 256, as in Speccy.

A larger window gives you narrower frequency bands and longer possible delays, but blurs rapid changes in the sound. A smaller window gives broader bands that follow changes in the input more closely.

Hover over WINDOW to see the details, for example Window: 256 samples, 128 bands; latency 6.7 ms.

Changing WINDOW clears the existing delays. The delayed sound fades out while the new analysis window fills, then fades back in. This takes a fraction of a second. The dry signal continues uninterrupted.

BAND sets how many adjacent frequency bands share a delay. The available settings are 1, 2, 4, 8 and 16. The default is 1, so every band has its own delay, producing the most detailed and smeared result.

Grouping bands keeps more of the sound's shape intact. At higher settings, the delays fall into wider frequency ranges, producing something closer to a bank of filtered echoes.

Hover over BAND to see the setting, for example Band: 4 bins share a delay.

SEED selects the delay pattern, from 1 to 50. The same seed produces the same pattern on every Re-Speccy, so you can reproduce a sound across instances. The default is 25.

Hover over SEED to see its number.

SPREAD controls how different the right channel's delays are from the left's. At zero, the two channels use the same pattern, so a mono input stays mono. Turn SPREAD up to give the right channel increasingly different delays. At full setting, each channel has its own independently generated pattern.

Changing SEED, BAND, SPREAD or SHAPE doesn't switch the pattern abruptly. The old delays fade into the new ones over two windows.

DELAY

TIME sets the maximum delay, from 1 window at the minimum to 128 windows at the maximum. The range changes evenly on a ratio scale, rather than in equal steps. Each band gets a delay somewhere between a quarter of a window and the maximum set by TIME.

TIME starts fully right, at 128 windows, which matches Speccy's range.

You can turn or modulate TIME while sound is passing through; the delays adjust smoothly.

Hover over TIME to see the current range in windows and seconds, for example Time: up to 128.0 windows, 0.68 s.

SHAPE controls how the delays are distributed across the frequency spectrum.

At the centre, marked RANDOM, the delays are distributed according to SEED, without any preference for high or low frequencies.

Turn SHAPE right to make delay times increase with frequency: the low frequencies arrive first and the high frequencies last. Turn it left to reverse the pattern, so the high frequencies arrive first.

At either end, the delays follow this pattern with a little random variation still present.

Hover over SHAPE to see the current setting, for example Shape: 60% rising, highs longest.

FEEDBACK sends each band's delayed signal back into its own delay, producing repeats at that band's own rate.

At minimum, you get one repeat. At maximum, each repeat retains 98% of the previous signal, so it can take a long time to fade away. FEEDBACK starts at zero.

Hover over FEEDBACK to see how much signal is retained on each repeat.

DAMP makes the high frequencies fade faster than the low frequencies in the feedback loop, much as echoes in a room tend to lose their brightness over time.

At maximum DAMP, the highest band loses three quarters of its level on each repeat, while the lowest band loses none. The bands in between fall somewhere between those extremes.

DAMP starts at zero and has no effect unless FEEDBACK is active.

Hover over DAMP to see how much level the highest band loses on each repeat, for example Damp: the top loses 45% a pass.

MODULATION

TIME, FEEDBACK and MIX each have a CV input and an AMT knob.

AMT is off at its centre position. Turn it right to let the incoming voltage move the control; at full right, a ±5 V signal can sweep the control across its full range. Turn AMT left of centre to reverse the effect.

The CV signal is added to the control's current knob setting.

FREEZE stops new audio entering the delay buffers. The sound already in them continues to repeat, with each band playing back at its own rate, for as long as FREEZE is active. This works regardless of the FEEDBACK setting.

Switch FREEZE on and the existing sound is held. Switch it off and new audio enters the delays again, with FEEDBACK returning to its normal behaviour.

You can operate FREEZE with its button or the GATE input. A voltage above 1 V turns freeze on; a voltage below 0.5 V releases it. Either the button or the gate can activate freeze.

I/O

IN L and IN R are the left and right inputs. If IN R is unpatched, IN L feeds both channels.

MIX blends the dry input with the delayed sound, from DRY to WET. The blend uses an equal-power curve, helping the overall level stay consistent as you turn the knob.

MIX starts at 0.2, so the output is mostly dry, as in Speccy.

At the DRY end, the output is the input signal without any added delay, provided it stays within ±5 V. See the output description below.

Hover over MIX to see the wet proportion.

OUT L and OUT R, on the cyan block, are the stereo outputs. Signals pass through unchanged up to ±5 V. Above that, the output gradually bends towards ±10 V and never exceeds it, preventing heavy feedback from producing unlimited output levels.

The delay buffers aren't saved with the patch. Loading a patch starts with empty delays. The control settings, including the FREEZE button state, are saved as usual.

The display

The display across the top shows the incoming and delayed sound as a spectrum, from 40 Hz on the left to 20 kHz on the right. Frequency is laid out on a logarithmic scale, across 48 display bands.

The coloured bars, running from violet to cyan, show the delayed sound and its repeats. A thin, light line shows the incoming sound being analysed.

The display uses a decibel scale, from 66 dB below 5 V at the bottom to 5 V at the top. It shows whichever of the two channels has the higher level.

Levels rise immediately when a sound arrives, then decay at about 30 dB per second. This keeps short sounds visible for a moment after they've finished.

Read the line and bars together to see what's happening. The line shows what's coming in; the bars show what's coming back out. With FEEDBACK turned up, the bars can remain visible after the input has faded. With FREEZE on, the bars hold their existing pattern while the line continues to follow the input.

Jacks

All unpatched inputs read 0 V.

Notes

Steady tones retain their level. The delay system accounts for each frequency band's phase when reading between windows. A sustained note therefore comes back at its original level, regardless of the TIME setting or how TIME moves.

The delays are repeatable. SEED determines the pattern; it doesn't change over time. To make the pattern move, modulate TIME or change SEED. Changes fade in rather than happening abruptly.

CPU use is steady. Re-Speccy spreads the work of analysing each window evenly across the samples in the next. Processing load is therefore much the same from sample to sample, without large spikes, at every WINDOW setting.

For Speccy users

• Stereo input and output. Re-Speccy has separate left and right inputs and outputs, with SPREAD to control the difference between their delay patterns. You no longer need two instances for stereo.
• New controls: TIME, SHAPE, DAMP and FREEZE. With TIME fully right and SHAPE centred on RANDOM, the delays cover the same range as Speccy and are distributed randomly.
• CV inputs. TIME, FEEDBACK and MIX can all be modulated, and FREEZE has a gate input.
• The dry signal is now genuinely dry. In Speccy, even the dry signal passed through the analysis system and acquired latency. In Re-Speccy, the dry signal bypasses that system entirely.
• No latency light or menu. There is no latency setting to adjust.
• Different seed patterns. Seed 25 in Re-Speccy produces a different delay pattern from seed 25 in Speccy.
• Feedback works per band. Each band repeats through its own delay, as in Speccy. The maximum feedback retains 98% of the signal on each repeat.

Starting points

A chord unfolding. Feed a sustained chord into IN L. Set WINDOW to 1024, BAND to 1, TIME about halfway, SHAPE to the right of centre, and MIX to three-quarters. The low notes arrive first, followed by the higher frequencies and harmonics.

Shimmering repeats. Feed in a short phrase. Set WINDOW to 512, FEEDBACK to about 70%, DAMP to a quarter, and MIX halfway. Each frequency band repeats on its own schedule, gradually losing its high frequencies. Change SEED to try a different pattern.

A held spectrum. Play into Re-Speccy with FEEDBACK turned up and MIX fully wet. Press FREEZE when the sound reaches the point you want to capture. The bands continue cycling through their delays. You can also patch a slow gate into FREEZE to capture and release sounds in turn.

Filtered taps. Set BAND to 16, WINDOW to 2048, TIME low and FEEDBACK about halfway. The delays fall into broad frequency groups, producing something like a bank of filtered echoes.

Wide from mono. Feed a mono source into IN L. Set SPREAD fully right, TIME low and MIX to about one-third. The two channels get different short delays, widening the sound without obvious echoes.

A moving pattern. Connect a slow LFO to TIME's CV input and set AMT to about a quarter. The delays gradually lengthen and shorten together, making the pattern breathe.