How to Tune Vocals So Nobody Can Tell
Vocal tuning · 5 min read
Correct where the note sits, not how the singer arrived at it. Leave the first 80–120 ms of every note untouched, keep the vibrato, and only move notes more than 20–25 cents off. That is the whole difference between tuning nobody hears and tuning everybody hears.
Leave this part alone
80–120 MS
The attack of every note · correct only past 25 cents
The 60-second answer
Use a graphical editor when the goal is invisible. Move the body of each wrong note and leave the attack, the slide into it and the vibrato as sung. Skip anything inside 20 cents. If you must use a real-time tuner: Retune Speed 40–90, Flex-Tune around 40, Humanize around 40.
The ear judges the middle of the note, not the start
Pitch perception needs time. On a sustained tone most listeners start noticing an error around 5–10 cents, but only after the note has been running for a moment. During the first tenth of a second the ear is busy with the consonant and the attack. That is why the start of a note can be 40 cents off and sound fine while the same 40 cents on the sustain sounds broken. Tuners correct from the first sample, removing the one part nobody was judging.
Scoops, falls and vibrato are the signature
Every singer has a way into a note: a scoop from a semitone below, a hard landing, an overshoot then a settle. That approach identifies a voice as much as its timbre. Flatten it and every note starts the same way, which is exactly what listeners hear as auto-tuned without knowing why. Same with vibrato: 4–7 cycles a second, usually 20–50 cents wide. Compress that and the note stops breathing.
The invisible tuning method, step by step
| # | Do this |
|---|---|
| 01 |
Listen once and mark only what is wrong Play the take against the beat and note the four to eight moments that actually bother you. That list is the job. Fix 4–8 notes, not 200 |
| 02 |
Set your threshold at 20–25 cents Under that, leave it. A quarter of a semitone is inside what a good singer does naturally and what a listener hears as expressive. Ignore under 20 cents |
| 03 |
Move the note body, not the whole blob Correct the sustained middle. If the tool moves the note as one object, split the attack off with a note separation first. Correct the sustain only |
| 04 |
Protect the first 80–120 ms Whatever the tool, the front of the note stays as sung. In a real-time plugin, Flex-Tune and Humanize buy you this. Attack untouched |
| 05 |
Keep 70–100% of the vibrato Reduce it only if it is genuinely wobbly, never zero it. A dead-flat held note is the loudest possible tell. Vibrato: 70–100% kept |
| 06 |
Formant correction on past 50 cents Big moves shift the resonances of the voice and make the singer sound like someone else. Then A/B against the untuned take at matched level. Formants locked past 50 cents |
The trick: leave the first 100 ms of every note completely untouched
This is the highest-return move in vocal tuning and almost nobody does it. Take a note that needs correcting and split it: the first roughly 100 ms — consonant, attack and slide into pitch — and everything after. Correct only the second part. The note now sits perfectly in tune from the moment the ear starts judging it, and it still arrives the way the singer sang it, scoop and all. In a graphical editor you do this with a note separation just after the attack, then move only the second blob. With a real-time tuner you approximate it by keeping Retune Speed at 40–90, which simply means the correction takes long enough that the attack has already passed. Either way the vocal measures in tune and reads as a performance. A/B it level-matched on one line: the version corrected only past 100 ms sounds like a better singer, the fully corrected one sounds like a plugin.
Stop guessing the chain.
The Vocal Chain Bible is 88 real vocal chains from Grammy-winning engineers — every plugin, every setting, in order, with free alternatives for each one. 5+ years, 100+ interviews, one PDF.
Get the Vocal Chain Bible — $48 →Mistakes that kill it
- Tuning every note because you can see them. A screen full of blobs is not a list of problems. If it did not bother you on playback, leave it.
- Snapping notes to the grid. Perfect equal temperament on a voice sounds mechanical against real instruments, which are never exactly in equal temperament either.
- Zeroing vibrato then adding a vibrato effect. You removed the singer's timing and replaced it with an LFO. It never matches.
- Tuning a doubled stack note for note the same way. Identical pitch on lead and double removes what made the stack sound wide.
Invisible tuning is a subtraction job. Decide what genuinely needs fixing, fix only that, and hand back everything else as performed.
Keep going
FAQ
How many cents off is actually noticeable?
On a sustained note, most listeners start hearing it around 10–15 cents and clearly hear it past 25. On short notes the tolerance is much wider.
Is graphical tuning always better than real-time?
For invisible correction, yes, because you choose which part of which note moves. Real-time is faster and correct when the tuning is meant to be heard.
Why does my tuned vocal sound thin?
Big pitch moves shift the formants with the pitch. Turn on formant correction, or re-sing anything more than a semitone out.
Who's behind this
Ivan Meshcheriakov · IvanFromRnD
5+ years ago I moved to LA with zero connections and started building from nothing — 14-hour days, meeting people and interning at studios. It took me 5+ years to meet and interview 100+ Grammy-winning engineers — then I turned it all into the Vocal Chain Bible.
Read my full story →