Contour / resonance and motion
Contour reworks the performance ideas in Kuwa Mashine’s nature instrument: resonant material, musical modes, a breathing bed and repeatable gestures. A noise-excited delay line forms the string, with damping, body resonance and stereo placement. The phrase runs against sample time inside an AudioWorklet, so drawing the picture does not set the timing.
Eightfold / relationships between voices
SOMA’s LYRA-8 manual describes eight freely tuned voices, paired relationships, envelope-dependent frequency modulation, a Hyper LFO, double delay and distortion. Our research included the official DIY archive and its labelled PCB/tweak drawing. That archive contains board layouts, parts and assembly material; it does not provide a complete standalone circuit schematic.
Eightfold implements eight triangle-to-soft-square voices, slow drift, envelope-weighted cross modulation and selectable pair, bank or ring connections. Two filtered cross-feedback delays precede drive. It is not a Schmitt-trigger or PT2399 circuit simulation. Differences from the hardware include digital envelopes, a different control layout and original nonlinear behaviour. Hardware component tolerances and the exact analogue saturation are not modelled.
Two useful software precedents are Mike Moreno’s LIRA-8 and Diego Capoccitti’s OktaDie. Both demonstrate ways to make the organismic idea playable in software. Neither supplies the source of this browser instrument. We did not verify a working online LYRA implementation during this research.
Attractor / expression inside the voice
Generate’s useful architectural lesson is a nonlinear generator feeding a wavefolder and low-pass gate, with expression preserved per note. Attractor uses original bounded coupled oscillators and phase interactions, a sinusoidal folding stage and per-voice filtering. Its three generator behaviours are original, not the named Generate algorithms. Aggressive high notes can produce digital aliasing; this is an exploratory instrument, not a substitute for Generate’s proprietary DSP.
Velocity sets note amplitude. Pressure changes each voice’s energy and instability. CC74 changes its colour and folding. Pitch bend stays on its member channel. Manager sustain is distributed across the zone; RPN 0 sets pitch range. Lower and upper MPE zones are available. Hardware-controller testing is a separate check from verifying MIDI message handling.
Repeats, shimmer and the input channel
Microcosm’s manual suggests a territory of rhythmic taps, captured fragments and overlapping pitch-shifted grains. Our effect reads real captured samples at twice their original rate through two overlapping windows. That octave layer enters filtered delay and diffusion. It is an original shimmer effect, not a model of a proprietary Microcosm algorithm.
External audio has its own trim, high-pass, compression, low-pass, saturation and effect sends. Its monitoring path is separate from its recording path. The master is captured as stereo PCM and encoded into an actual 24-bit WAV. Session JSON stores settings separately, keeping the distinction between a patch and a recording clear.
Beat machine / one clock, four parts and a voice
The fourth instrument runs its melodic and drum lanes in one AudioWorklet processor. Sixteenth-note boundaries use the audio sample counter, including paired swing intervals. Pattern changes are queued until step one. Browser animation reports the audio clock’s position and does not trigger the notes.
Harp uses decaying harmonic modes with per-note expression. Bass combines a polyBLEP saw and sine sub with glide and a one-pole filter. Kick, snare, hat and metallic percussion are synthesised; no third-party drum recordings are bundled. Drone uses detuned tones and slow phase motion. Four stereo outputs feed independent dry paths and sends into a shared effect return. The voice uses a separate input-processing rack. Mute and solo affect the recorded mix.
Live events enter at the next available audio render quantum rather than waiting for a JavaScript sequence timer. The context requests interactive latency. Input/output buffers, the device driver, dynamics processing and wireless output can still add delay. The displayed base/output figures are browser estimates, not a measured round-trip result.
Timing and headroom
Synthesis, phrase generation, pitch-shifting and recording run on the browser’s audio rendering thread. Output starts attenuated, delay feedback is bounded and soft-saturated, and a final compressor reduces peaks. This does not guarantee protection from every interface or speaker configuration. Base latency is shown by the browser; controller-to-ear and microphone round-trip latency depend on hardware, drivers and the whole route.