Voice changer
What is a vocoder?
Speak into a vocoder and what comes out is your sentence, pronounced by a synthesizer. The words, the rhythm, and the phrasing are yours; the tone belongs to the machine. Here is how one signal ends up wearing the shape of another, and where the robot voice actually comes from.
The vocoder is the oldest trick in the voice changer family, and it is still the most recognizable: the talking robot, the singing machine, the synthesizer that pronounces words. Unlike a pitch shifter, it does not bend the voice you already have. It measures your voice and uses the measurement to control a completely different sound. Once you see that split, everything about how a vocoder behaves, including its famous limitations, follows directly.
Two signals: the carrier and the modulator
Every vocoder takes two inputs. The modulator is the signal that carries the message, almost always a voice: it decides what gets said, when, and how the energy moves as you talk. The carrier is the signal that supplies the tone, usually a synthesizer waveform rich in harmonics, like a saw. On its own the carrier is a steady, meaningless drone. The vocoder's job is to imprint the modulator's shape onto the carrier, so the drone starts talking.
The mechanism is a bank of frequency bands. The vocoder splits both signals into the same set of bands, low to high, and continuously measures how loud the modulator is in each one. Each measurement then drives a volume control on the matching band of the carrier. Where your voice has energy, that carrier band opens; where your voice is silent, it closes. Do that across every band at once, dozens of times a second, and the carrier's spectrum is forced into the silhouette of your speech. Your mouth is sculpting the synth.
The analysis bank and the synthesis bank
In textbook terms, the band idea is two filterbanks running in parallel. The analysis bank points at your voice: a stack of band-pass filters, each one followed by an envelope follower that tracks how much energy its slice of the spectrum holds from moment to moment. The follower's attack and release times decide how quickly each band reacts, fast enough to catch consonants, slow enough not to flutter. The synthesis bank points at the carrier: the same stack of band-pass filters, but each one feeding a gain stage instead of a detector.
The wiring between the two banks is the whole machine. Band by band, the analysis side's envelope becomes the synthesis side's gain:
out_k(t) = carrier_k(t) * env_k(t)Band k of the output is band k of the carrier, multiplied by the loudness the analysis bank just measured in band k of your voice. Run 16 of those in parallel and sum them, and the carrier speaks. It is the same standard DSP everyone has used for decades, and it is exactly what runs in patchd's engine.
The band count is the resolution of the imprint. Few bands means each volume control governs a wide stripe of spectrum, so the output is coarse, heavily robotic, and harder to understand. More bands means a finer copy of your vocal shape, more intelligible and closer to speech, at the cost of some of the machine character. That is why vocoders traditionally come in sizes: 8 bands is a growl, 16 is the classic balance, 32 starts to sound like a synthesizer with excellent diction. Since each band is just filters and multiplies, the same arithmetic every DSP effect is made of, a vocoder runs in real time with no meaningful delay.
Why it sounds like a robot
Here is the detail most explanations skip: the output's pitch is the carrier's pitch, not yours. The analysis bank measures how loud each band is, but it never asks what note you are singing, so your intonation is discarded at the door. If the carrier is a fixed monotone, the output is your words delivered on one unwavering pitch, and a voice with perfect articulation and zero melody is precisely what we call robotic. The deadpan is not a flaw in the machine; it is the machine.
The second signature is what happens to consonants. Sounds like s, t, and f are bursts of noise, not tones, and a pitched carrier simply has no noise in it to shape, which is why crude vocoders sound like they are mumbling through a kazoo. Real designs patch the hole in two ways: they let a little of the voice's own high-frequency sibilance pass straight through, or they blend some noise into the carrier so the synthesis bank has raw material for the consonants. Both knobs exist on any serious vocoder, including the one in patchd.
Where it came from
The vocoder was not invented for music. It was built in the late 1930s at a telephone research laboratory as a compression scheme: band envelopes are far cheaper to transmit than the voice itself, so speech could be squeezed down long-distance lines and rebuilt on the far end. During the Second World War the same idea, with the envelopes encrypted in transit, carried secure voice traffic between allied leaders. Decades later musicians discovered the reconstruction step was more interesting than the compression: feed the synthesis bank a synthesizer chord instead of a bland test tone and the telephone machine sings. By the late 1970s the robot voice had moved from the switchboard to the charts, and it has never really left.
The Vocoder node in patchd
In patchd, the vocoder is not a metaphor; it is a node. Vocoder lives in the Voice FX section of the rack's + Add Node menu, part of the Studio tier, and its panel is the textbook machine with real controls. Carrier sets the internal carrier's pitch, 50 to 400 Hz with a default of 110: lower is a deeper robot, higher a thinner one. Bands is the resolution switch, 8, 12, 16, 24, or 32, with 16 as the balanced default. Mix blends the vocoded signal against your dry voice, and at 0 percent the node is a clean passthrough.
The rest of the panel is the character drawer. Tone swaps the carrier's waveform across ten flavors, from the bright Supersaw default through Square, Pulse, Pad, Organ, String, Bell, Choir, Brass, and Sub, and Chord can harmonize the carrier into a major, minor, fifth, or octave voicing so the robot sings in harmony. The consonant fixes from the previous section are here too, by name: Sibilance (default 40 percent) passes your s and t sounds through, while Noise and Noise Type blend white, pink, brown, breath, or vinyl noise into the carrier. Per-band reaction speed is the real Attack and Release pair, 3 ms and 35 ms by default.
Because it is an ordinary rack node, the vocoder sits in your chain like any other effect: after your cleanup, before your output, on any strip. And because it is pure DSP, it belongs to the instant family of voice effects rather than the AI family, a distinction that matters once latency does; the honest comparison lives in DSP vs AI voice changing.
The robot, on tap
A vocoder is a measurement wired to a multiplication: an analysis bank that reads your voice band by band, and a synthesis bank that forces a carrier to follow. That is the whole secret, and it is why the effect has survived ninety years of audio technology, it is simple, it is instant, and nothing else sounds like it. In patchd the Vocoder node puts that machine one + Add Node click away, alongside the rest of the Voice FX family and Persona itself. patchd is in development now. Join the waitlist and we will tell you the moment it is ready to install.