Fractus: origin story

Here’s the complete DSP summary — the whole signal path, written up plainly, if you want to see exactly how the thing is built under the hood.

What it is: a time-domain granular processor extending the Mutable Instruments Clouds concept, with much higher grain polyphony (up to 256 voices vs. Clouds’ Cortex-M4 limits). Clean-room DSP — no Clouds/Parasites source referenced or ported.

Signal path:

buffer → grain engine → [Grain Mix: dry rejoins here] → Character → diffuser → reverb → Output Trim → output

1. Capture / buffer stage

• Stereo ring buffer, 32-bit float, host sample rate. Allocated once in prepareToPlay at 30s × the real sample rate — no reallocation while audio runs.
• Buffer Length (seconds or bars/beats-synced) moves a read/write window inside that fixed allocation.
• Freeze locks the buffer as a static grain source. The frozen buffer travels with saved sessions and presets as embedded float audio, written only when frozen.
• Captured audio is treated as content, not transient DSP state — it survives a prepareToPlay reconfigure (e.g. host buffer-size change) unless the allocation itself must reshape (sample-rate change).

2. Grain engine — the core differentiator

• Fixed-size voice pool (up to 256, default 32), each voice a read position + envelope phase + pitch/rate + pan/spatial position + amplitude + lifetime.
• Scheduler, three modes: free-running internal clock, host-tempo sync (phase-locked to the host’s ppq, note-division quantized), and a genuine Poisson (exponential inter-arrival) random/stochastic mode.
• Envelope: boxcar / triangle / Hann, continuously morphable via a precomputed table.
• Grain Size: 1ms–4000ms, a resource-free parameter — memory and worst-case CPU are unaffected by grain length; only overlap, bounded by the voice ceiling, matters.
• Confinement: a grain is kept inside the recorded region for its whole life, not just at spawn; where it can’t fit, it’s shortened rather than allowed to wrap and click.
• Reverse: 0–100% per-grain probability of playing backward, reading the same chunk of audio entered from the far end — so it sounds like backward tape, not a mirrored read.
• Per-grain spatial placement: Pan (field center) + Scatter (randomization around it), continuous rather than a mode enum; grains reflect rather than clamp at the field edges.
• Loudness compensation: 1/√overlap, latched per grain at spawn from its own delivered length and rate — keeps Density/Grain Size/Max Grains from doubling as volume controls, with no zipper artifacts.
• Voice stealing: steals only from voices past ~70% through their envelope, biased to minimize audible disruption.

3. Diffuser
Multi-stage all-pass network smoothing grain-edge artifacts, with dry/wet, channel crossfeed, feedback (contained to the diffuser/reverb loop — never routes back into the live grain buffer), and reverb send.

4. Reverb — clean-room, written from scratch
Three algorithms: plate, FDN (feedback delay network), and early-reflection. Fully modulation-matrix-assignable (type, decay, pre-delay, tone/damping, size). Wet level is normalized against decay time, so Decay is a length control rather than a volume control.

5. Character macro — the “quality” control
A post-buffer, non-destructive lo-fi/hi-fi macro spanning saturation, band-limiting, sample-rate decimation, and bit-crushing, each with its own attenuator (a unidirectional 0–1 depth scaler, eurorack-style). At Character 0 the whole chain is bit-exact identity. Loudness-compensated so the macro doesn’t double as a volume knob. Each stage’s taper is mapped to the audible range of that specific effect.

6. Modulation matrix
Control-rate (not audio-rate) routing: 8 slots, sources → destinations, applied as an offset on top of the host parameter’s base value — never overwriting the parameter itself, so it never fights DAW automation. Sources: 4 multi-shape LFOs (tempo-syncable), an audio-rate envelope follower on the plugin’s input, 2 random/S&H generators with slew, and the MIDI note envelope. ~26 destinations across nearly every exposed parameter.

7. MIDI

• Note mode: Off / Pitch (drone, notes retune the cloud) / Pitch+Gate (monophonic synth with ADSR VCA).
• Polyphony: 16-voice, one grain cloud with per-grain pitch rather than N separate engines — density scales with held-note count so a chord gets denser rather than louder.
• MIDI Learn: right-click any control to map a CC, which takes over the parameter’s full range and writes host automation like a mouse drag.

Precision: Float32 for the capture buffer and grain engine (memory-bound); double for filter/delay-line state in the diffuser and reverb, where accumulated error actually matters.

Here’s a short history of how granular processing evolved and influenced us

Here’s where the idea behind it actually comes from….  the history of granular and our influences!

Where “grains of sound” actually comes from

The idea is older than you’d think, and it didn’t start in a music studio.

In 1947, a physicist named Dennis Gabor — the same Gabor who later won a Nobel Prize for inventing holography — proposed that sound could be described as a series of tiny acoustic “quanta,” little grains of energy in time and frequency, borrowed straight out of quantum physics. He wasn’t thinking about music at all. But the idea that any sound is really just a pile of grains, and could be reassembled into something else entirely, planted a seed that took decades to grow into something you could actually play.

Iannis Xenakis is the one who dragged it into music. A composer, an architect, and a genuinely strange and brilliant thinker, Xenakis started treating sound as clouds of discrete events instead of continuous lines — closer to describing a swarm or a weather system than a melody. That’s a key part of this story by the way… that’s where the name Fractus comes from. A “fractus cloud” is a ragged, broken one, and “broken into pieces and reassembled” is precisely what this plugin does to whatever you feed it.

Curtis Roads is the one who wrote the book — literally, Microsound, still the closest thing granular synthesis has to a bible. He spent decades turning what had mostly been instinct and improvisation into an actual vocabulary the rest of us could use.

And Barry Truax is the one who made it real-time. For most of its history, granular synthesis lived on paper and in overnight computer runs — you’d render a piece and come back the next day to hear it. Truax built systems that could generate grains live, as you played, which is the difference between a theory and an instrument.

None of that lineage belongs to any one company or module or designer. It’s decades of physics, composition theory, and stubborn engineering — and it’s the reason Fractus’s About box actually credits Gabor, Xenakis, Roads and Truax by name, instead of just saying “inspired by Clouds” and calling it a day.

Why Clouds?

Which brings me back to the module. In 2015, Mutable Instruments — designed by Émilie Gillet — released Clouds, and it did something a lot of granular hardware before it hadn’t managed: it made granular synthesis fun. Not just powerful, not just theoretically interesting… fun to play. Lofi ambient beds built from a single frozen string ensemble. Sound design where a door creak became a creature voice. Live performances, where hitting “freeze” with tempo-sync grains made a groove. Clouds invited experiments.