Deluge Owlet Firmware (1.3 Port)
[!NOTE] Day-to-day development happens elsewhere. New feature work lives in
owlet-labs/DelugeFirmware-talon. This repository still receives releases and bug fixes, and is where tagged builds are published.
This is owlet-firmware, a personal fork of the Deluge Community Firmware maintained at owlet-labs/DelugeFirmware. It serves as a playground for experimental sound design features that may be too specialized or CPU-intensive for the main community branch.
This branch is based on main for stability.
Rationale/use case
Most of the features in this branch are built to support my personal use case for the deluge:
- A portable inspiration station/sound design machine and companion for external gear (useful effects added and lookback sampler make sampling external gear into loops and resampling very easy.)
- Inspiring moreso than granular control over complex effects.
- Less focus on raw voice count more focus on inspiring sound design and related workflow
- This being said effects are designed to have minimal footprint when disabled except for binary size. i.e. dont allocate memory dont consume cpu when switched off. With everything "off" the baseline perf/voice count etc. should be ideally close to that of the normal deluge firmware.
Status
Highly experimental and for most users, not a replacement for the main community firmware. Folks interested are encouraged to experiment, but we recommend backing up sd cards just in case.
- owlet-firmware-1.3 is the default branch and has a stable baseline.
- dev branch tracks development work as it lands here. https://github.com/owlet-labs/DelugeFirmware/blob/dev/README.md
Known limitations
- Horizontal menus must be enabled - the UI for these effects requires horizontal menus to function properly. Future work will seek to decouple this requirement.
- Disperser (OWLPASS) has high CPU usage with >8 stages
- Sine shaper (HOOT) is moderately CPU hungry
- Other effects are fairly light
Features (vs Main)
Workflow
| Feature | Description |
|---|---|
| Bounce in Place | Render a synth/kit/audio clip (or a whole track) through its current FX chain into a new AudioClip on a new adjacent AudioOutput. Source clip and output remain intact. Menu: clip context menu → "Bounce Clip" (single) or "Bounce Track" (every clip on the source output). Synth/audio sources record pre-reverb (MIX channel) so reverb is recreated live on the bounced clip; kits bake per-drum reverb into the WAV but preserve the kit-wide send as a live send (a single clip send can't model per-drum variation). Reverb-send automation is cloned onto the new clip via AutoParam::cloneFrom, preserving curves. Sample clusters are pre-primed before the render so audio-clip and sample-based kit-drum sources don't record leading silence while SD loads. Playback window is trimmed to exactly clipLength-in-samples so AudioClip's timestretch engine runs at ratio 1.0 (no mid-clip artifacts). Gain stage set to unity (overriding the AudioClip 25% default) so the bounced clip plays at source level. |
| Retrospective Sampler | Lookback buffer for capturing audio after the fact. Sources: Input, Master mix, or Focused Track. Duration modes: time-based (5/15/30/60 sec) or bar-synced (1/2/4 bars). Bar modes tag filenames with BPM. See testing guide. |
| Sound Cloning (Grid/Kit) | Clone sounds via pad gestures. Grid view: hold a clip pad + press an empty column to clone the synth track (sound + notes). Kit view: hold an occupied audition pad + press an empty row to clone the drum sound. Cloned sounds are independent copies with unique names for reliable save/reload. |
| Record Source Selection | Choose the recording source for Record to Sample via Sound Editor → Actions → Record source. Options: Left input (auto-detect, default), Stereo input, Balanced input, Deluge mix (no fx), Deluge output (fx). Enables resampling the Deluge's own output directly into a sample slot. Setting persists across power cycles. |
Effects
| Feature | Description |
|---|---|
| Scatter (Bird Brain) | Beat-repeat, slice manipulation effect and granular looper. Rate knob controls slice length. Grain mode with dual-voice crossfade. pWrite and density are patchable params. See testing guide. |
| Featherverb | Lightweight 4-tap FDN/allpass cascade reverb. 77 KB buffer vs Mutable's 128 KB. See testing guide. |
| Multiband OTT Compressor (DOTT) | 3-band upward/downward compressor with CHARACTER, RATIO, VIBE, and SKEW controls. 8 vibe zones for dynamic character. Can function as full downward, full upward, or anywhere in between. Aggressively optimized. Includes band-level metering. See testing guide. |
| Sine Shaper (HOOT) | Harmonic waveshaping with drive, harmonic content, symmetry, and mix controls. Adds musical saturation and overtones. See testing guide. |
| Table Shaper (PELLET) | XY wavetable-based waveshaper with smooth interpolation between shapes. See testing guide. |
| Disperser (OWLPASS) | Allpass cascade effect with zone-based topology. Available per-voice and in GlobalEffectable chain. High CPU usage with >8 stages. |
| Automodulator | Automodulation effect with phi triangle routing and 4-point XY LFO wavetable. Available per-voice and in GlobalEffectable chain. |
| Eroder | Noise-modulated delay line comb filter with IIR feedback. SVF-filtered noise (bandpass) + zero-crossing sample-and-hold + pitched triangle modulator drive a short delay line. Phi triangle banks on freq (delay, feedback, soft clip, fold, HPF cutoff, pitched mod depth) and character (depth, resonance, S&H blend, width, pitched mod offset). Pitch tracking on comb, SVF, HPF, and triangle. Q31 integer waveshaping (soft clip + triangle fold). Available per-voice and in GlobalEffectable chain. |
| Utility Effect | Channel-strip utility with four controls: Volume (−∞ to +12dB gain staging), Pan L and Pan R (independent per-channel panning, 0=hard left, 127=hard right), and Stereo Width (0%=mono, 100%=normal, 200%=exaggerated). Placed before DOTT in the signal chain. Available on synth tracks, kit rows, audio clips, and master. All params at default = zero CPU (bypass). |
| Harm Effect | Harmonic rebuild tool: note-tracking notch filter (pre-reverb) strips a frequency, sine oscillator (post-reverb) adds it back clean. 102 harmonic ratios (denominators 1/2/3/4/8, numerators to 32) for precise interval selection. Notch filter is a biquad with width control (surgical to wide). Oscillator has AR envelope, portamento, phase catchup (click-free retrigger over ~24 cycles), and pink noise output scaling (−3dB/oct, calibrated at 40Hz). Level and Fine Tune are mod matrix targets (hybrid patched params). Pitch bend tracking. Last-note-priority pitch with fallback to held voices. Sound-only (synth tracks and kit rows). ~3200 cycles/buffer (0.28% CPU). |
Synthesis
| Feature | Description |
|---|---|
| Pulse Width Triangle | New oscillator type (TrianglePW / TRPW). Triangle wave with dead-zone pulse width control — PW knob compresses the triangle into a narrower region with silence in the dead zone. Waveform visualization in pulse width menu. |
| Phi Morph Oscillator | Procedural oscillator generating waveforms from phi-triangle banks. Two zone knobs (A/B) each produce a distinct 8-segment waveform shape. Crossfade (Wave Position) morphs between them with morph excitation effects (amplitude overshoot, curvature boost, phase distortion). Waveform shaping via sine blend, odd symmetry, windowing, and slope matching. Per-sample modifiers: phase jitter, amplitude-dependent noise, asymmetric gain for even harmonics. Supports pulse width modulation. |
| Phi Weave Oscillator | Scanned-physics oscillator (Mathews/Verplank scanned synthesis, ICMC 2000): a closed ring of 32 masses on springs simulated at control rate and scanned as a seam-free wavetable at audio rate — pitch is the scan speed, timbre is the physics. Zone knobs (A/B) select the laws of motion via phi-triangle banks: per-node stiffness/damping/coupling landscapes, the home shape the ring relaxes toward, a continuous bow, note-on pluck shape, and slow ring rotation. One shared string state per Sound: the A/B crossfade interpolates physics under continuous state (click-free by construction), and the crossfade's own motion bows the string — sweeping Wave Position is a playing gesture. Cheaper per sample than PhiMorph. See physics deep dive and testing guide. |
| Phi Vox Oscillator | VOSIM voice oscillator (Kaegi & Tempelaars, Institute of Sonology — JAES 1978): each cycle emits a burst of raised-cosine pulses per formant branch, decay-scaled, then silence — two branches make a vowel, and formants are fixed in Hz so vocal identity survives transposition. Zones anchor on eight voice characters (Breath…Rasp) with phi-triangle wander far beyond vocal physiology (growing-pulse rasp, alternating-polarity hollowness). The A/B crossfade is a diphthong, and crossfade motion injects voiced-gated consonant bursts — the wave index knob makes it speak. A Formant Track menu item (0–50) blends fixed-Hz formants toward note-relative ratios (harmonic-locked overtone behavior), a slow φ-triangle shifts both formants coherently across zone space (the same vowel findable higher/lower), and formant frequencies glide with ~20ms portamento. Stateless per voice (zero buffers, ~260 bytes per Sound). See VOSIM design notes and testing guide. |
| Phi Swarm Oscillator | Injection-locking / synchronization oscillator (Adler 1946, Kuramoto 1975): the note's master phase drives two Adler-coupled slave oscillators whose behavior traces the Arnold-tongue map — locked to exact rational ratios inside a tongue (self-tuning consonance), injection-pulling "sirens" at the edges, never-repeating quasiperiodic beating outside, with Langevin temperature melting the network from crystalline through analog drift into noise. The wave-index crossfade ANNEALS: knob motion heats the swarm out of lock, parking it lets you hear it re-crystallize. Zones Still…Chaos; pitch-invariant locking; per-voice state is two phase words in an otherwise-unused field. See sync physics and testing guide. |
| Phi Gendy Oscillator | Xenakis dynamic stochastic synthesis (GENDYN lineage — Formalized Music; Serra CMJ 1993; Hoffmann CMJ 2000) with locked pitch: a 16-breakpoint polygon whose amplitudes perform the GENDYN double random walk between elastic barriers, once per buffer, scanned at the note pitch. Phi-triangle landscapes paint per-node entropy and cage terrain (one waveform region boils while another stays calm); a home-shape spring gives calm zones a timbre to relax into; crossfade motion STARTLES the walkers and the spring reels them home. Zones Haze…Frenzy trace an exponential entropy arc from barely-alive warmth to reflection-grit. Rich saw-class spectra, PHI_MORPH-class cost, zero per-voice state. See design notes and testing guide. |
| Phi Stair Oscillator | Staircase wavetable (stepped-waveform lineage: Walsh-function synthesis — Chamberlin 1980, Hutchins' Electronotes): up to 16 steps whose heights follow zone pattern families (Brick, Terrace, Mesa, φ-random Glyph, Teeth…), widths carry asymmetry landscapes, and a slope dimension sweeps hard-edged pulse-class brightness into ramped mellowness. Step count is continuous (fractional last step — sweeps never pop), loudness is statically normalized per morph position, and the stereo zone knob works via dual staircase taps. Zero per-voice state. See design notes and testing guide. |
| Phi Stereo Zones | Zone-style stereo control shared by the phi oscillators (Stereo item in the oscillator menu): position within each of eight zones sets the stereo amount (every zone starts mono), the zone selects the character — PhiWeave/PhiStair dual pickup taps (distance, tonal tilt, counter-scan), PhiGendy lagged polygon taps (~3ms micro-slap on walk events), PhiSwarm slave separation with beat-opposition autopan. Zones Slim…Vast. Works with unison off; composes with unison stereo spread. PhiVox excluded. |
| Unison Index Modulation Source | New per-voice modulation source (UNISON_INDEX) that assigns a unique value to each unison voice. Patch it to any oscillator-level parameter (pitch, volume, phase width, wave index, phase, carrier/modulator feedback, start offset) for per-voice timbral variation from a single note. Three knobs in the Unison menu control the distribution: Shape (8 zones: Linear, Power, S-Curve, Step, Triangle, Sine, Random, Drift), Mapping (8 zones: Symmetric, Anti-sym, Pitch+, Pitch−, Center-out, Rotate, Pairs, Shuffle), and Curve (bipolar warp from center-heavy to edge-heavy). Drift shape provides per-block bounded random walk for evolving textures. All zone knobs have intra-zone meta parameters for fine control. |
Sampling
| Feature | Description |
|---|---|
| Gate Sample Mode | New sample repeat mode for drum kit rows. Notes act as mute/unmute gates — the sample loops for the duration of the note. If BPM is detected in the filename (e.g. _120BPM or bare _120_), the sample is tempo-synced and phase-locked so triggering mid-bar starts at the position it would have reached if playing since the downbeat. Without BPM in the filename the sample still gates normally, just without time-stretching or phase-lock. Gate View Editor: Press SCALE in kit mode to enter gate view (SCALE LED lights, "GATE" popup). Green pads = audible, red pads = muted. Tap to toggle. Hold first + tap second on same row to span-fill. Hold a green pad + twist vertical encoder to set per-segment attack (fade-in, 0-127), horizontal encoder for release (fade-out). Hold multiple green pads simultaneously to adjust all at once. Exponential time scaling: 0 = instant, ~64 = 50ms, ~96 = 400ms, 127 = ~2.4s. Values are saved per gate segment and serialized with the song. Also available for audio clips. |
| Loop Crossfade | True dual-read crossfade at loop boundaries. Configurable duration (0–1000 ms) per source in the sample menu. When approaching the loop end, a second read from the loop start fades in while the main read fades out — seamless blending with no volume dip. Works from the first loop iteration by reading the fade-in from the sample cache while the uncached path fades out. For split-loop offset wrapping, time-stretch mode, and pingpong loops, falls back to fade-out/fade-in envelopes (still smoother than no crossfade). |
| Loop Pingpong | New PINGPONG sample repeat mode. Playback bounces between loop start and end points instead of jumping back. Each unison voice tracks its own direction independently. After note-off while going backward, playback continues to loop start then stops. |
| Plockable Sample Start Offset | Patchable (plockable) sample start offset — varies per-step via parameter locks. Offset applied as a fraction of the sample-start-to-loop-end region. Split-loop architecture: when offset crosses the loop start, the voice plays offset→loop-end first, then loops normally from loop start. Works with timestretch, reverse, and pingpong. offsetWraps toggle wraps past loop-end back to sample start for full-range modulation. Live offset updates during playback for looping modes (repitched and time-stretch). |
Modulation
| Feature | Description |
|---|---|
| Macros (MACRO 1–4) | Four global modulation sources that are also patchable parameters. Assign any mod source to a Macro, then patch the Macro to multiple destinations — enables modulation-of-modulation (e.g. LFO→Macro→Filter+Pitch). One-frame evaluation delay prevents feedback. Self-routing is blocked. Available in the mod matrix as both sources and destinations. Default value 50% (bipolar center). |
MIDI
| Feature | Description |
|---|---|
| MIDI Follow: Mod Knob CC | Map 16 consecutive MIDI CCs to the Deluge's 16 mod knob slots (8 pages × 2 knobs). Menu: MIDI → MIDI Follow → Mod Knob CC. Set a base CC number (1–112); CCs from base to base+15 control mod knobs in order. Supports CC learn. Value 0 = OFF. Persisted in flash storage. |
| MIDI Follow: CC Presets | Switchable CC mapping presets. Drop .XML mapping files into SETTINGS/MIDIFollow/ on the SD card. Menu: MIDI → MIDI Follow → CC Preset — scroll through available presets and select to load. A Standard.XML preset is auto-created from firmware defaults on first boot. Selected preset is persisted to SETTINGS/MIDIFollow.XML for reboot survival. Community can share controller-specific mapping files. |
Developer
| Feature | Description |
|---|---|
| FX Benchmarking Framework | Performance profiling tools for DSP development. |
For Developers
DSP Documentation:
- Phi Triangle Parameter Evolution - Design pattern for multi-parameter control evolution using golden ratio frequencies
- Phi Weave Physics - Scanned-synthesis oscillator: model, stability analysis, and morph-bow energetics, with simulation figures and original literature
- Phi Vox VOSIM Notes - VOSIM voice oscillator: pulse-burst formant synthesis, statelessness, and diphthong morphing, with figures and original literature
- Phi Swarm Sync Physics - Injection locking, Arnold tongues measured from the firmware's own update rule, and the annealing morph, with figures and original literature
- Phi Gendy Stochastic Notes - Xenakis dynamic stochastic synthesis: the double random walk, elastic barriers, phi terrain and the startle morph, with figures and original literature
- Phi Stair Step Notes - Staircase wavetable: pattern families, the slope dimension, continuous step count, and the Walsh-function lineage, with figures
This fork is open source under the same GPL-3.0 license as the community firmware. Feel free to:
- Fork this repo for your own experiments
- Cherry-pick individual features into your own builds
- Submit PRs to the main community repo if you refine something worth sharing
No guarantees of stability or compatibility. Use at your own risk.
Dear Deluge owners
If you want to start using the community firmware please visit https://delugecommunity.com/ for all relevant details.
About
The Deluge from Synthstrom Audible is a portable sequencer, synthesizer and sampler. Synthstrom Audible decided to open source the firmware application that runs on the Deluge to allow the community to customize and extend the capabilities of the device.
The hardware is built around a Renesas RZ/A1L processor with an Arm® Cortex®-A9 core running at 400MHz and 3MB of on-chip SRAM. Connected to that is a 64MB SDRAM chip, a PIC for button handling and either a 7-Segment array or an OLED screen. The application is written in C and C++ and some occasional assembly in bare metal style without an operating system.
Important links
- For acquiring and using the firmware visit the community firmware website
- Once you have the firmware, consult our detailed update guide for instructions on how to install it
- To contribute to the project with code, bug reports, suggestions, and feedback see: How to contribute to Deluge Firmware
- To learn about the toolchain, application, and to start working on the code continue with: Getting Started with Deluge Development
- Please also visit the Synthstrom Forum and the community Discord to interact with other users and the developers
- You can donate to the project using the Synthstrom Patreon