tongues: a 20 HP vocal chopper, by daylight noise

Take

Hold TAKE, the button or a gate at the TAKE jack, and TONGUES records IN, up to 8 s, in place of the take it had. It listens as it records, following the voice's loudness and the moments its sound changes and tracking its pitch to tell what's voiced from what isn't, so when you let go the cut is done a moment later. It cuts in the dip between two vowels, on the consonant that starts the next one.

Silences are dropped. A syllable is 70 to 600 ms long (a longer one is split at its deepest dip), and a take holds 32 of them at most. While TAKE is held you hear what it's taking at the outs; the old take keeps singing until the new one is cut, and a press too short to hold a syllable leaves it alone.

Sing

GATE: each rising edge sings one syllable, rising in 3 ms. When the gate falls it lets go, in 12 ms. Four voices, so a syllable letting go overlaps the next.
V/OCT: the note it's sung at, 0 V at middle C. The syllable's own pitch, its glides and leaps with it, is moved onto that note and held there, so the chops land in tune.
What's voiced is moved a pitch period at a time: each of the voice's pulses, with the ringing after it, laid down one new period apart. What isn't, an s, an f, a t, a breath, plays as it was. So the consonants come through the move intact.

Word and order

WORD, the big knob: which syllable. Its travel walks the take from the first syllable to the last. WORD CV adds 0 to 5 V across the same span, so a sequencer there picks syllables exactly.
ORDER all the way down: stuck. Every gate sings WORD's syllable again, a stutter.
At noon: in order. Each gate takes the next syllable, round to the first after the last, so the take is sung back in the order it was said, to your melody. Turning WORD moves the place it reads from.
All the way up: shuffled. Every gate a syllable at random, never the same one twice running. Between noon and the top the chance of a jump grows; between the bottom and noon, the chance of moving on.

Cut

CUT is how much of each syllable plays: 15 ms at the bottom, just its consonant, a click; about 95 ms at noon; all of it at the top. It's taken when the gate rises, so turning it changes the next syllable, not the one sounding.

If the gate is still high when it gets there, it holds: the pitch period at that point is repeated, a period of the note apart, for as long as the gate lasts. A short CUT under a long gate holds a buzzing consonant. A full CUT holds the vowel, since a syllable holds at the end of its vowel at the latest.

Mouth and spread

MOUTH: the size of the mouth, apart from the pitch. At noon it's the voice's own. Up, smaller, the formants as much as an octave higher: a child, then something smaller. Down, bigger, as much as an octave lower. The note stays put. MOUTH CV adds a volt for every 12 semitones.
SPREAD: each syllable its own place in the stereo field. All the way down, every one in the middle; toward noon, alternating left and right, wider as it turns; past noon, more and more of them thrown somewhere at random, out toward the edges, until at the top every syllable lands hard left or hard right.

The lights

Sixteen windows in a row, lit from behind in a dusty rose, are the take laid out as a strip. Each is a sixteenth of it, dim where there's voice and dark in its silences, and bright under the syllable being sung. While TAKE is held they fill from the left as the take grows, one for each half second of the eight.

The panel

Twenty HP, every part on the line's staggered grid and mirrored about the middle. The art is an open, singing mouth in gold dots on black, the face from the nose down. The lights sit in a row just behind the upper teeth, and WORD lies in the dark of the mouth, where the tongue would be. The windows are bare circuit board, pale in room light. TAKE is a lit button at the foot; its cap glows red while it takes.

IN: the voice it takes. TAKE: a gate in, the same as holding the button.
GATE and V/OCT: the notes. WORD CV and MOUTH CV: into WORD and MOUTH.
OUT L and OUT R: the syllables, each where SPREAD puts it. With OUT R unpatched, OUT L carries both.

Inside

The TONGUES panel singing, its knobs on.

Drag to look around. Lift the panel: the baffle stays on the board, a black block with a cell for each window, its LED at the foot. One RP2350 does it all: one core sings, the other cuts each take as it comes in, into 8 MB of memory beside it.

Specs

Format

Width: 20 HP
Depth: about 21 mm behind the panel, its power header included
+12 V: about 115 mA, 175 mA at the most (the strip at its cap, TAKE lit)
−12 V: about 10 mA
+5 V: not used; 10-pin header
Panel: 1.6 mm FR4: ENIG gold through black mask, white labels, sixteen windows of bare board
Controls: 5 knobs, WORD's big; TAKE, a lit button

I/O

IN: audio, ±5 V with room to ±9; 100 k
TAKE: gate, high past about 1.1 V; 100 k
GATE: gate, the same
V/OCT: 1 V an octave, 0 V at middle C, −2 to +8 V
WORD CV: 0 to 5 V across the take; 100 k
MOUTH CV: ±1 V for ±12 semitones; 100 k
OUT L, OUT R: ±5 V at most, 1 k (a take that peaks at 5 V sings at about 2.3); OUT R summed into OUT L while unpatched

Engine

Brain: RP2350, both cores, 46.875 kHz, 32-bit float
Converters: PCM1808 in, PCM5102A out, on one I2S clock
Memory: 8 MB PSRAM: the take, the one before it and the next
Voices: four, pitch-synchronous
Light: 16 RGB LEDs behind bare board, their current capped at 100 mA
Firmware: over USB-C or the debug header, on the back

Nothing's for sale yet.

TONGUES is designed and its firmware builds. Nothing has been made, so the numbers here are the design's and the engine's, not measurements, and there's no date.

hello@daylightnoise.com