model MK-1 · atomic audio division
A real channel vocoder, live in your browser. Your microphone (or a dropped-in WAV/MP3 file) is the modulator; your computer keyboard plays a fully polyphonic carrier — hold down a chord and the vocoder sings it. Twelve bandpass channels analyze your voice and reshape the carrier's tone to match, the same trick behind every robot voice and talkbox solo since the 1970s, running on nothing but oscillators and filters.
transport
modulator source
Mic uses your voice as the modulator. File plays a WAV/MP3 through the vocoder instead — no mic needed, and it loops while powered on. Drop a file anywhere on this panel to load it.
analysis display
channel bank (12 voiced + 1 unvoiced)
controls
keyboard
Two rows, like a piano: ZSXDCVGBHNJM,L.;/ is the lower
octave, Q2W3ER5T6Y7UI9O0P is one octave up. Hold
several keys at once for a chord — it's polyphonic. No mic? Switch
the modulator source to file above and load a WAV/MP3
instead — the keys above are clickable/tappable too, but the
vocoder needs a live mic or a loaded file to have anything to
shape.
How it works: the mic signal is split into 12 logarithmically spaced bandpass bands (160Hz–5.2kHz). Each band is rectified and smoothed into a slow-moving envelope — how loud your voice is, in that frequency range, right now. That envelope is fed straight into the matching band of the carrier (your keyboard notes, summed together so chords work for free), so the carrier's spectrum tracks your voice's spectrum in real time. A 13th band runs the same trick on filtered noise instead of your keyboard tone, so "s" and "sh" sounds get a bit of hiss instead of going silent.
No samples, no AudioWorklet, no server — just BiquadFilterNode,
WaveShaperNode and GainNode, synthesized live
in your browser.