Why your ears stop working on your own take
After the fifth pass on a vocal, you are no longer listening to it. You are remembering it. Your brain has learned the melody as you sang it, including the errors, so the wrong note starts sounding like the intended note. This is not a failure of talent, it is how auditory memory works, and it affects experienced engineers exactly as much as beginners.
There is a second trap. Intonation errors are relative, so a note that is 30 cents flat can sound perfectly acceptable on its own and obviously wrong the moment the piano enters. Judging a vocal solo tells you very little about how it will sit in the arrangement.
This is why intonation is one of the few musical questions worth answering with a number instead of an opinion. Not because the number is more musical, but because it does not get tired.
Cents: the unit that settles the argument
A cent is one hundredth of a semitone. There are 100 cents between two adjacent keys on a piano, which makes the unit fine enough to describe differences far below what most people can name but well above what they can feel.
Useful reference points:
- 5 to 15 cents: the range a professional singer typically measures. Perceived as in tune.
- Around 25 cents: the point where errors on sustained notes start being noticed by ordinary listeners.
- 50 cents: a quarter tone, exactly halfway between two notes. Unambiguously out of tune.
- 100 cents: a full semitone, which is a wrong note rather than an intonation problem.
Nobody hears in cents, and that is the point. The unit exists so that a disagreement about whether a take is acceptable can be resolved by looking at the same figure instead of trading impressions.
What counts as in tune
Two figures matter, and they answer different questions.
Average deviation describes the take as a whole. Under 10 cents is tight, professional intonation. Between 10 and 25 is normal for a good performance with human character. Above 25 the take reads as loose, and correction or a re-record becomes the honest conversation.
Off-pitch moments describes the exceptions. A take can average a respectable 12 cents and still contain four sustained notes sitting 40 cents flat, because averages hide outliers. Those individual moments are what a listener actually notices, so their count and position matter more than the average.
Concrete action: read both. A low average with several bad moments means punching in a few phrases. A high average with no standout errors means the whole take drifts, which usually points at the key or at monitoring rather than at the singer.
Only sustained notes count
This is the distinction that separates a useful measurement from a useless one. Sung melody is not a sequence of fixed frequencies: it slides, scoops into notes from below, bends at the end of phrases and uses vibrato that swings either side of the target by design.
A naive measurement counts all of that as error and reports a disaster on a take that is stylistically perfect. Soul, blues and most contemporary pop phrasing would score terribly, because approaching a note from underneath is the entire point.
A meaningful analysis therefore looks only at notes that are held and steady, where the singer clearly intended to land and stay. A scoop into a note is expression. Sitting 40 cents flat for a whole bar is not.
Concrete action: when reviewing flagged moments, listen to each one in context before correcting it. If the note was approached deliberately and left deliberately, it is phrasing. If it was aimed at and missed, it is a fix.
Sharp or flat: what each one tells you
The direction of the error is diagnostic, and it points at different causes.
- Consistently flat usually means the singer is running out of support: fatigue, insufficient breath, or a key that sits at the top of the range. It tends to get worse as the session goes on.
- Consistently sharp usually means pushing: too much effort, often from monitoring that is too quiet, so the singer overcompensates to hear themselves.
- Flat only on high notes points at the arrangement. The melody is above the comfortable part of the range.
- Random in both directions points at monitoring or at unfamiliarity with the melody rather than at technique.
Concrete action: if the take drifts flat towards the end, stop correcting and re-record the last third earlier in the next session. Tuning software can move the pitch but it cannot restore the energy that was missing from the performance.
When the key is the problem, not the singer
Before blaming technique, check where the melody actually sits. Two measurements make this concrete: the full range from lowest to highest note, and the tessitura, which is the narrower band where the voice spends most of its time.
A singer can reach a high note occasionally and still be unable to live there for a whole chorus. When the tessitura sits at the top of the range, intonation degrades, strain appears, and the take gets worse with every repetition rather than better.
Concrete action: if the flagged moments cluster at the top of the range, transposing the song down by one or two semitones will fix more than any amount of editing. It is the single most underused solution to a vocal that will not come together.
How much correction is too much
Pitch correction is a normal part of modern production and there is nothing to apologise for. What people react badly to is not correction itself but the artefacts of heavy correction: notes that snap instead of moving, transitions that jump instantly from one pitch to the next, a delivery that has lost its microtiming.
Two signatures give it away. The first is the proportion of time spent exactly on the grid: above roughly 70 percent starts to look unusual for a human performance, and above 85 percent is very difficult to achieve by singing. The second is the share of transitions that happen abruptly rather than gliding, which rises sharply when correction is set to retune fast.
Concrete action: if you are using correction, slow the retune speed until the transitions breathe again, and correct individual problem notes rather than processing the whole take. Both changes preserve the performance while fixing the errors that actually bother listeners.