Autotune a vocal
Pulls a sung or played line onto the notes of a key. Set it gentle and it is invisible tuning; set it instant and full and it is the hard-tuned effect. One voice or instrument at a time, and nothing is uploaded.
drop a vocal or instrument take here
or click to pick. A dry, solo recording works best — not a full mix.
| Pitch found | — |
|---|---|
| How far out it was | — |
| Correction applied | — |
The two controls that decide how it sounds
Everything people mean by "autotune" comes down to two numbers, and they do completely different things.
Strength is how far toward the correct note each moment is pulled. At 100% every pitch lands exactly on a scale note. Below that the singer's own pitch is preserved proportionally, so 35% tidies a wobble without erasing the fact that a person sang it.
Retune speed is how quickly the correction is allowed to move, and it is the one that creates the famous effect. Real singing slides between notes; every scoop into a note and every fall out of one is a glide. A slow retune lets those glides through and only fixes where a note settles, which is why it sounds like nothing was done. An instant retune forbids gliding entirely, so the pitch jumps from note to note in hard steps — that stepping is the hard-tuned sound, not a side effect of it.
For invisible tuning: 100% strength, natural or slow retune. For the effect: 100% strength, instant retune, and a pentatonic or minor scale so the jumps are wide enough to hear.
Pick the right key, or do not bother
Correcting to the wrong key is worse than not correcting. If a singer is slightly flat in G major and you tune them to C major, every note gets dragged to a pitch that was never in the song, and the result is confidently, precisely wrong.
If you do not know the key, use chromatic. It snaps to the nearest semitone regardless of key, which fixes tuning without imposing a scale, and it is the safe default for a part you did not write.
Fewer notes means a stronger effect. Chromatic barely moves anything; major moves more; pentatonic moves a lot, because whole notes are missing and anything landing on one gets pushed to a neighbour. That is exactly why pentatonic is the classic choice for the hard-tuned sound.
How the pitch is moved
Time-domain PSOLA. The recording is cut into short grains, each centred on one cycle of the waveform, and reassembled with the spacing between grains changed — closer together raises the pitch, further apart lowers it. The grains themselves are untouched, so the formants that make a voice recognisably that person mostly stay where they are.
The naive alternative is to resample, which moves everything including the formants and gives you the chipmunk effect. That is the difference between a tuned vocal and a sped-up one.
Grain placement is the part that decides whether this sounds clean or buzzy. Each grain has to be cut at the same point in every cycle, and picking the loudest sample does not achieve that on a bright voice, because there are several peaks inside one period. Marks here are placed on a low-passed copy where the fundamental has one clear peak per cycle. Getting that wrong was measurable: grain spacing wandered by 15% and the output dropped two octaves.
What to feed it
- One voice, dry. Reverb and delay smear the pitch, and the tail of a reverb reads as a note that was never sung.
- No harmonies or doubles. Two pitches at once means there is no single pitch to correct, and the result will chase whichever is louder moment to moment.
- Not a full mix. Use the stem splitter to lift the vocal out first if that is what you have.
- Trim the noise. Breaths and handling noise can read as very low, very quiet notes; cutting the dead air first removes most of them.
What it is not
This is pitch correction, not pitch editing. It moves notes onto a grid; it will not let you drag one note somewhere else, redraw a melody, or fix timing. That is a DAW job — Melodyne, Logic's Flex Pitch, Ableton's warping — and no browser page substitutes for it.
It also cannot rescue a take that is badly out. Correction of more than about a semitone starts to sound processed no matter how gently it is applied, because the grains are being stretched further than the format comfortably allows. If the report says the median note was more than 45 cents out, the honest fix is another take.
And if what you want is to move the whole recording to a different key rather than tune it, that is the pitch shifter.
FAQ
Is this real autotune?
It is pitch correction working the same way: detect the pitch, snap it to the nearest note of a key, and move the audio there with PSOLA. It is not the Antares plugin, and it does not do note-by-note editing like Melodyne.
How do I get the T-Pain sound?
Instant retune speed, 100% strength, and a pentatonic or minor scale. The hard-tuned effect comes from forbidding the glide between notes, so the pitch jumps in steps.
How do I make it sound natural?
100% strength with a natural or slow retune. That fixes where notes settle while letting the singer scoop into them, which is what makes correction inaudible.
Is my recording uploaded?
No. Pitch tracking and correction both run in your browser tab, which matters for an unreleased vocal.
What if I do not know the key?
Use chromatic. It snaps to the nearest semitone without imposing a scale, so it fixes tuning without dragging notes somewhere the song never went.
Will it make my voice sound like a chipmunk?
No. It moves pitch with PSOLA, which leaves the formants roughly in place. Resampling is what causes the chipmunk effect, and that is not what this does.
Can I autotune a whole song?
Not a full mix — there is no single pitch to correct. Isolate the vocal with the stem splitter first.
Does it change the timing or length?
No. The output is exactly the same length as the input, by construction.
Why does my result sound processed?
Probably because the take was a long way out. Corrections beyond about a semitone stretch the grains far enough to hear. The report tells you how far out the median note was.
Does it work on instruments?
Yes, on monophonic ones — a bassline, a lead synth, a flute, a single guitar line. Not on chords.