Voice changer
What is formant shifting?
Formants are the resonance peaks of your vocal tract, and they, not pitch, are what tell a listener how big the person talking is. Formant shifting slides those peaks up or down while the note stays exactly where it was, which is how the same voice becomes a giant or a sprite without changing pitch at all.
Open any serious voice changer and you will find a formant dial sitting next to the pitch dial, usually with no explanation of what it actually moves. The short version: formants are the resonance peaks of your vocal tract, the acoustic fingerprint of the tube your voice travels through, and shifting them changes how big the speaker sounds while the note stays put. This page is the long version: what a formant physically is, why sliding one transforms a voice's apparent size and character, what it sounds like when you push too far, and what the Formant Shifter node in patchd does, dial by dial.
A voice is a buzz plus a filter
Every voiced sound you make starts as a buzz. Your vocal folds open and close somewhere around 85 to 255 times per second, and that vibration rate is your pitch: the note your voice sits on. The buzz is not a pure tone; it carries energy at every whole-number multiple of that rate, a comb of harmonics marching up the spectrum at perfectly even spacing.
The buzz then travels through your throat, mouth, and nasal passages, and those cavities behave like a filter. Frequencies the tract resonates at come out louder; the rest come out quieter. The peaks of that filter, the frequencies your vocal tract amplifies, are the formants, numbered from the bottom up: F1, F2, F3, and so on.
The first two do a job you use constantly without noticing: they carry vowel identity. When your tongue and jaw move from "ee" to "ah", you are not changing pitch; you are reshaping the tract, which moves F1 and F2 to a new pair of positions, and your ear reads that pair as the vowel. Same buzz, different filter, different vowel. This is the textbook source-filter model of speech, and it splits a voice into exactly the two properties a voice changer can grab independently: the harmonic comb, which is pitch, and the resonance envelope, which is the formants.
What a formant shifter actually moves
A formant shifter separates those two properties and moves only the second one. The harmonic comb passes through untouched, so the note stays where it was; the resonance envelope is lifted off the voice and re-fit onto it higher or lower than it came in. In patchd the shift is measured in semitones, and the move is a single multiplication:
f_shifted = f * 2^(st / 12)Every resonance peak is multiplied by the same factor. At -3 st the factor is 2^(-3/12), about 0.84: F1 at 700 Hz lands near 589 Hz, F2 at 1100 near 925, F3 at 2500 near 2102. Pitch never appears in the formula, which is the whole point.
Why does sliding the envelope change who the voice sounds like? Because formant positions are set by the physical length and shape of the vocal tract, and your ear has spent a lifetime learning the correlation: a longer tube resonates lower, and a longer tube belongs to a bigger body. Shift the envelope down and the same note suddenly arrives from a larger chest, deeper and roomier without being one semitone lower. Shift it up and the voice sounds smaller, brighter, lighter-built. The judgment is involuntary; you cannot hear a voice without sizing its owner.
The figure above also shows why pitch survives the trip: pitch lives in the spacing of the harmonics, and the shifter never touches the comb. That is what separates formant shifting from the old speed-up-the-tape trick, which drags pitch and formants together and lands on the classic chipmunk. The two dials, and when to reach for each, get their own page in formant shift vs pitch shift.
What it sounds like when you push too far
Small moves are surprisingly natural. A semitone or two either way reads as a slightly different person, and most listeners could not tell you what changed. Push past a few semitones and the math keeps working while the plausibility quits, because the shifted envelope now describes a vocal tract no human body contains. Vowels drift first: F1 and F2 carry vowel identity as a pair, so slide them far enough and one vowel starts wearing a neighboring vowel's coordinates, and speech gets subtly harder to parse. The character of consonants goes next, taking on a cupped-hands hollowness. A big downward shift turns into talking out of a barrel; a big upward one goes insect-thin.
The fix, when you want a big transformation rather than a subtle one, is not more formant. It is moving formant and pitch together, in agreement, so the note and the apparent body size tell the same story. That pairing, and why every convincing deep or bright preset is built from it, is covered in deep and female voice changers, explained.
The Formant Shifter node in patchd
In patchd the effect is its own rack node. Formant Shifter is one of the Studio voice-fx nodes, and its panel is honestly small. The main dial is Formant, running -12 to +12 semitones with 0 as untouched. Mix is a dry/wet blend from 0 to 100 percent, defaulting to 100, and 0 is a bit-exact copy of the dry signal, not an approximation. Quality trades analysis detail against responsiveness, worth raising on low, deep voices where the envelope is harder to track, and Order sets how much vocal-tract detail the analysis resolves: higher finds more formant peaks and more character, lower is smoother and softer.
Under the hood it is a source-filter method, the same model this page teaches: the node estimates the envelope, peels it off, slides it by 2^(st/12), and re-fits it, all in the time domain with no FFT block to wait for. That is why the node adds no latency of its own; it just rides the normal engine path, so it is safe to leave in a live chain while you monitor yourself talking.
Its neighbor in the same palette, the Pitch Shifter, carries both dials at once: Pitch from -24 to +24 semitones plus its own independent Formant dial from -12 to +12, so one node can move the note and the size separately or together. Which node to grab, and why the two moves are independent at all, is the subject of formant shift vs pitch shift.
Why your EQ cannot do this
It is tempting to file formant shifting next to EQ, since both are about frequency, but they are different machines. An EQ applies a fixed gain at a fixed frequency: a bell at 700 Hz makes whatever happens to pass 700 Hz louder, on every vowel, forever. A formant is a moving target, F1 and F2 relocate every time your articulation changes, and a static boost cannot follow them, let alone relocate them. EQ reshapes tone, brighter, less boomy, more present; it never changes the apparent size of the instrument.
The same goes for the de-esser, which is a compressor aimed at the sibilance band: it turns harsh "ess" sounds down when they spike, but turning a band down is still gain, not relocation. None of the cleanup nodes move formants, and the Formant Shifter moves nothing else. Tone repair and character change are different jobs, and patchd keeps them in different nodes.
Where you will meet formants next
Formants are also the reason Persona's instant voices work at conversation speed. The DSP voices in patchd's voice changer are built from exactly the moves on this page, pitch and formant shifts plus character processing, so they add almost no delay on top of your normal mic path; the AI voices go further and replace the voice outright, for about 350 ms of added delay. All of them are original characters we built, never imitations of a real person, and every voice carries its own keybind, including one for your real, unprocessed voice. The Free tier includes 6 rotating taster voices, with the launch lineup targeted at about 25. patchd and Persona are in development now. Join the waitlist and we will tell you the moment they are ready.