A dual-deck crossfaded overdub looper built on monome's softcut-lib. Each deck holds two softcut voices (a Rev-swapped pair), so there are four voices in all. Unlike the buffer-tape shuttle/tape engines, softcut's read/write head plays and records the same loop at once, with subsample-accurate click-free loop crossfades and interpolated overdub - layering live input onto a running loop with a feedback control. That sound-on-sound overdub is the reason to reach for softcut over the existing tape engines.

Four voices is the hardware-measured safe CPU budget (worst case ~62% avg / 79% peak). See docs/dev/softcut-spike.md for the feasibility measurements and the path to a future 6-voice build.

Controls (focused track)

Each deck edits one focused voice at a time; the Rev pad swaps which one, and the platform reseeds the pots so the absolute knobs catch the newly-focused voice without a jump. Both voices of a deck keep sounding.

Control	Function
PITCH	rate, bipolar about noon: noon = stop, CW = forward to +2x, CCW = reverse to -2x
POS	loop start (slides the loop window through the buffer)
SIZE	loop length (a short stutter up to the whole buffer)
MIX	track volume
ENV	overdub feedback: 1.0 = infinite sound-on-sound layering, lower = old layers decay as you add, 0 = overwrite (only acts while overdub is armed)
Alt+POS	pan
Alt+PITCH	SD loop slot - loads a clip into the loop buffer (boot-preloads slot 1-2 per deck)
MOD_AMT	loop crossfade time - short = tight, long = lush/smeared loops
MODFREQ	rate slew time (glide on varispeed sweeps)
FLUX pad + PITCH	post-filter cutoff (boots open; sweep down to engage)
FLUX pad + MIX	post-filter resonance

Pads

Play - roll / stop the focused loop (snaps rate to unity on engage).

Alt+Play - the record gesture, context-sensitive on whether the voice has content:

Empty voice (looper record): first press starts recording a fresh loop (the buffer is cleared first); the second press closes the loop at exactly the length you played and starts looping it. Further presses overdub.

Voice with content (SD-loaded or already recorded): arms / disarms overdub on top (sound-on-sound).

Rev - swap the deck's focused voice.

Seq - realign all voices to their loop start at once: softcut's cutToPos does a click-free crossfaded jump, so drifted free-running loops snap back to a common downbeat (the voice-sync gesture).

Alt+Seq (tap) - save the full take (the whole recording, lossless) to the focused voice's currently-selected SD slot.

Alt+Rev - save the trimmed loop (exactly the POS/SIZE window - "bounce what you hear") to the same slot.

Alt+Seq (hold ~1.5 s) - erase the focused voice (clean discard - nothing is written). It empties the buffer so the next Alt+Play records a fresh take. Holding cancels the tap's pending save, so an erase never touches the card.

Both saves write a float32-mono WAV; pick the destination slot with Alt+PITCH first (use an empty slot so you don't overwrite a boot loop). The ring flashes amber while it writes (~1 s, non-blocking - the loop keeps playing). Save and load share the same slot files, so a captured loop reloads on the next boot or via Alt+PITCH. Alt+Seq is the safe everyday save (SIZE/POS stay non-destructive); Alt+Rev prints exactly the loop you hear. Note the Alt+Seq save commits a moment (~1.5 s) after the tap so a hold can turn it into an erase instead; Alt+Rev saves immediately.

The audio input is summed to mono (both input jacks), so a loop records whichever input the source is patched to.

Routing switch (L / C / R)

Per-voice pan into the stereo bus: LEFT (DoubleMono) manual Alt+POS pan; CENTRE (Stereo) auto-spread a deck's two voices hard L/R; RIGHT (GenerativeStereo) random pans, re-rolled on entry.

Make a loop

On an empty voice, Alt+Play to start recording, play your phrase, Alt+Play again to close the loop - it now repeats at the length you played.

Alt+Play once more to overdub. Keep layering (ENV high) or let old layers fade as you add (ENV lower).

PITCH for varispeed/reverse, the FLUX pad to filter, Seq to re-sync against the other deck.

Each voice's buffer is ~10.9 s (2^19 frames @ 48 kHz, 2 MB), so a single take can be up to ~10.9 s at the default 1x record speed.

Build / flash

make ENGINE=softcut            # normal SRAM build (-O2, ~89% SRAM_EXEC)
make ENGINE=softcut METER=1    # + on-device CPU load meter over USB serial / ring A
make program-dfu               # flash (enter DFU first)

Loop clips are 32-bit float, mono, 48 kHz WAVs at softcut/loop_<a|b>_<1..8>.wav on the SD card - the same format Alt+Seq/Alt+Rev save, so saved loops reload directly.

Not in v1

Phase-quantised voice sync - the Seq-pad realign covers the musical case; locked phase-quant is a later add.

6 voices - gated on the std::function-per-sample removal in the vendored core (see the spike doc).