What is autotune?

Strictly speaking, Auto-Tune is a product name that escaped into the language, the way velcro did. What everyone actually means by autotune is pitch correction: software that hears the note you sang, works out how far it sits from the note you meant, and closes the gap. That one mechanism has two completely different careers. Run gently, it is the invisible safety net under a huge share of the vocals you hear every day. Run at full speed, it is an instrument of its own: the hard, robotic snap that a whole era of pop and R&B was built on. This page explains the mechanism, then the knob that flips it from one career to the other.

Pitch detection, then a snap

Every pitch corrector does three things in a loop, hundreds of times a second. First it measures: a pitch tracker estimates the fundamental frequency of your voice right now. Then it decides: it compares that pitch against a grid of allowed notes and picks the nearest one. Then it moves: a pitch shifter transposes your voice by the difference. The grid itself is not mysterious. Western tuning spaces every note a fixed ratio apart:

Which rungs count as allowed is set by the key and scale. A chromatic setting allows all twelve notes in every octave, so the corrector simply snaps to the nearest semitone. Pick a real scale, say A major, and the grid thins out to the seven notes that belong, so every correction lands somewhere musical. One more thing a good corrector does quietly: it moves your pitch without moving your formants, the vocal-tract resonances that make your voice sound like you. That is why correction does not chipmunk you the way naive pitch shifting does.

Retune speed is the character dial

Here is the part that decides everything. The corrector knows where your pitch should go; retune speed sets how fast it travels there. Give it a few hundred milliseconds and the correction can only chase slow drift, so your vibrato, scoops, and slides all survive. Set it to zero and your pitch is welded to the grid the instant the tracker reports a note. Same algorithm, opposite sound:

That right-hand panel is the sound you already know. Late-2000s hip-hop and R&B turned it into a signature, hyperpop built a whole genre on top of it, and most people search for it as the T-Pain effect: a voice that jumps note to note with no glide in between, warbling wherever the melody moves. Nothing in that sound is a malfunction. It is retune speed at zero, chosen on purpose, and it only comes alive on sung, sustained vowels with movement in them. Flat talking gives the corrector nothing to snap, which is why speaking through hard autotune mostly sounds like nothing happened.

Key and scale decide where it can land

Chromatic correction is the safe default: the nearest semitone is never more than half a semitone away, so the corrector can never make things dramatically worse. But it also cannot make them very interesting. If you drift more than fifty cents toward a note that is not in your song's key, chromatic snapping will happily land you on it, perfectly in tune with a note that does not belong.

A scale setting fixes that by shrinking the target list. With the key set to A and the scale set to major, the corrector can only land on the seven notes of A major, so every snap is at worst an odd melody choice, never a sour pitch. This matters double for hard-tune: the iconic effect is essentially always run on a real scale, because chromatic hard-tune staircases through passing semitones and stops sounding like music. The flip side is that the key has to be right. Correct a song in D against an A major grid and the corrector will yank notes out of the song's own key with total confidence. When in doubt, pentatonic scales are the forgiving option: five notes, wide gaps, very hard to land anywhere ugly.

The Autotune node in patchd

What correction costs

Pitch shifting cannot happen one sample at a time. The shifter needs a window of audio to resynthesize, and the window is a real delay, which is why Autotune is the heaviest effect in patchd's rack and why the app is honest about it. The node's Quality setting states its own price: Low Latency runs about a 26 ms resynth and is the default, the only setting that fits a live monitoring chain; Balanced doubles that for a smoother result; High doubles it again and belongs on recordings, not conversations. The Hard engine runs a touch lower than Natural, one small mercy of its rigid approach. And because a pitch tracker will happily chase breath and consonants, a Voicing control sets how confident the tracker must be before it corrects at all, so silence and whispers do not send the corrector hunting garbage.

Autotune, vocoder, or voice changer?

Three effects get tangled together in search results, and the untangling is short. Autotune corrects the pitch of your voice; even at its most robotic, the timbre is still you. A vocoder replaces your voice with a synthesizer that wears your speech like a filter, which is a different robot entirely. And a voice changer transforms who you sound like altogether: in patchd that is Persona, which layers original DSP and AI character voices at the end of the chain, past the whole rack. Autotune is the one of the three that belongs to music first, and it earns its place the moment you sing a held note into it. patchd is in development now. Join the waitlist and we will tell you the moment it is ready to install.