Auto-Tune is a product, not a technique
Auto-Tune is a specific piece of software, made and trademarked by Antares Audio Technologies, that corrects the pitch of a recorded vocal. It is not the name of the process. The process is called pitch correction, or vocal tuning, and a great many products do it — Auto-Tune was simply the first one that worked well enough to become famous, and its name stuck to the whole category the way Hoover stuck to vacuum cleaners.
That distinction matters more than it sounds, because most of what people mean when they say “autotune” is not pitch correction at all. They mean one particular extreme setting of it — the hard, stepped, synthetic sound on records by Cher, T-Pain and a thousand imitators. Ordinary pitch correction, used the way most records use it, is completely inaudible, and it is on far more music than you would guess.
So there are really three separate things sharing one word: the company’s product, the general technique, and a famous effect that the product can produce but usually isn’t asked to. The mechanism behind all three is the same, and it is worth understanding once.
Seismic data, and then a Cher record
Auto-Tune was invented by Andy Hildebrand, a research engineer who had spent his career in the oil industry working on seismic data — sending sound into the ground and interpreting the reflections to find deposits. The mathematical tool for that job is autocorrelation: sliding a signal against itself to find the interval at which it best repeats.
A repeating interval in a reflected shockwave tells you about rock. The same measurement on a human voice tells you the pitch, because pitch is the rate at which a waveform repeats. Hildebrand realised the technique transferred, and Antares released Auto-Tune in 1997.
The intent was transparency — fix a slightly flat note, leave everything else alone. What made it famous was somebody turning it all the way up. Cher’s “Believe” in 1998 used the most extreme setting of the retune control, producing a voice that snapped between exact pitches with no human slide between them. The producers initially attributed the sound to a vocoder pedal. It was Auto-Tune, and within a decade T-Pain had built a career on the same setting and Kanye West had made an album out of it.
A tool built to be unnoticeable became famous for being extremely noticeable, which is why the word now means the opposite of what the software was for.
What it does to a vocal
Three steps, and they are the same three in every product of this kind. The software works out what pitch the voice is singing at each instant; it decides what pitch that note ought to be; and it re-renders the audio at the corrected pitch without changing the timing.
The first and third steps are solved engineering problems. The middle step is a musical judgement, and it is where nearly every bad-sounding tuned vocal comes from. Left on its default, the software targets the nearest note of a chromatic scale — all twelve semitones available at all times. If a singer aiming for a C lands slightly flat, the nearest chromatic note may be a B, and the software will confidently pull the note to B: now firmly in tune with nothing, and far more wrong than the original flatness was.
Telling it the key of the song removes that entire failure mode. The full walk-through of the three steps covers scale selection, chord tracks, and why formant preservation is what stops a shifted note sounding like a smaller person.
The robotic effect comes from one control
This is the single most useful thing to know about these tools, and it is almost universally misunderstood: the famous sound is not the result of tuning a vocal too hard. It is the result of tuning it too fast.
Two different controls are involved. Strength — how far a note moves toward its target. Retune speed — how quickly it gets there once the software has decided to move it.
A human voice never arrives at a pitch instantly. It slides in from below, overshoots slightly, settles, and adds vibrato. Those transitions carry most of the expression in a phrase; the portamento between two notes is often the most human thing in a vocal.
Set retune speed slow and the software waits through the slide, correcting only the sustained middle of the note. Nobody hears anything except a singer who is in tune. Set retune speed to zero and every transition is quantised — the voice jumps between exact pitches with no slide at all. Because human voices physically cannot do that, the ear identifies it immediately as a machine.
Which means strong correction with a slow retune is transparent, and gentle correction with an instant retune is unmistakably processed. If a tuned vocal sounds robotic and you did not want it to, retune speed is the control to change, not strength.
Automatic and manual are different jobs
Products in this category generally offer two ways of working, and they suit different problems.
- Automatic. You set a key and a scale, the software corrects everything in real time against it. Fast, useful for a rough vocal or for the deliberate effect, and indiscriminate — it will “fix” a blue note, a deliberate bend and a scoop into a note along with the genuine mistakes.
- Manual or graphical. The software draws the pitch curve it detected and you edit it note by note, moving only what is wrong. Slower, and the only way to tune a performance without flattening the parts of it that were intentional.
Most professional tuning is the second kind, applied to four or five notes in a take rather than to the whole thing. If a tool will not show you the pitch curve, you are trusting its note segmentation blindly — and segmentation is exactly where it will misread melisma. Euphona’s vocal tuning returns the detected curve as an editable track for that reason, with a slow retune as the default rather than a fast one.
The rest of the category
Auto-Tune is one product among many, and naming the category after it obscures some real differences in approach.
- Celemony Melodyne takes the graphical approach furthest, and can separate individual notes inside a chord — a genuinely different capability from correcting one monophonic line.
- Tools built into the DAW. Logic, Cubase and most other major hosts ship a pitch editor. For fixing a handful of notes these are usually sufficient, and you already own them.
- Alignment tools handle the adjacent problem of timing — matching a double or a harmony to a lead take rather than to a grid, which is generally what a double is for.
- Model-based tuners, including ours, use a learned pitch tracker rather than autocorrelation. The difference shows up on the hard material — breathy singing, growl, the first milliseconds of a note — where classical tracking makes octave errors. They also tend to be metered rather than licensed, so the comparison against a perpetual licence is a question of how much you tune rather than a flat one.
None of these is Auto-Tune, and none of them should be called it. They are pitch correctors, and Auto-Tune is Antares’.
Getting the effect deliberately
The hard-tuned sound is a legitimate production choice with a thirty-year history, and it is easy to get wrong by aiming at the wrong control.
- Retune speed to zero. This is the effect. Everything else is secondary.
- Set a narrow scale. The stepping is only audible if the steps are far apart. A chromatic scale gives the voice twelve nearby targets and the jumps become small enough to lose; the major or minor scale of the song gives it seven, and a pentatonic fewer still. The effect gets stronger as you remove notes.
- Sing straighter than feels natural. The effect reads best on a performance with little vibrato and deliberate, wide interval jumps, because the software has something obvious to snap.
- Keep it dry, or process it after. Reverb printed before the tuner confuses the pitch tracker. Tune first, then treat — and if the take arrived with room on it already, dereverb belongs before the tuner, not after.
The effect is a mix decision as much as a vocal one, and it behaves differently once the track is finished — hard-tuned vocals are dense and bright, and what a master can still change about that is less than people expect.
Tuning so that nobody notices
The opposite goal, and the one most records actually want.
- Start from the best take. Correction fixes pitch. It does not fix phrasing, timing, tone or conviction, and those are what make a vocal good — most of what goes wrong at the microphone cannot be corrected later at all.
- Always set the key and scale. The chromatic default causes most bad results.
- Slow the retune unless you specifically want the effect.
- Correct the notes that are wrong. Most takes have four or five that need it; most software will happily process all of them.
- Protect vibrato. Vibrato is deliberate pitch modulation, and a fast retune reads it as error and flattens it.
- Judge it in the mix. A soloed vocal reveals every artefact; a vocal in the track reveals the ones anyone will actually hear. If the vocal is fighting the arrangement rather than the tuning, that is an EQ problem, not a pitch one.
- Keep the untouched take. A week later you will want to compare, and you will sometimes prefer the original.
One case deserves its own warning: tuning a vocal that is still inside a finished stereo mix. The tracker has every other instrument to contend with and will follow the loudest pitched thing it can find. Separating the vocal out first makes it tractable — though a vocal extracted and re-summed is never quite the same as the original take, for the reasons separation’s own limits describe. If you have the multitrack, use it. If you only want to hear whether the split is clean enough to be worth tuning, the free two-stem vocal remover answers that in a couple of minutes without an account.
Autotune, auto-tune, Auto-Tune
All three spellings are in common use and they do not mean quite the same thing. The product is Auto-Tune, capitalised and hyphenated, and it belongs to Antares. Autotune, one word and lowercase, is what most people type when they mean either the effect or the general practice. Auto-tune as a verb means to apply the process.
We use “pitch correction” and “vocal tuning” for the technique and call our own tool a vocal tuner, because those are the accurate terms and the brand name is not ours to take. If you arrived here searching for the other word, this is the thing you were looking for — and the mechanism article is the longer answer to how it works.
Keep reading