These steps use DropWave Tune, a real-time pitch correction plugin for one vocal (AU, VST3 and CLAP on a Mac, free during the beta). The ideas are the same in any pitch correction.
1. Pick the key first
Correction pulls each note to the nearest note of a key, so the key decides everything. Pick it and its scale. Not sure? Press Detect: it listens to the vocal, and to the beat if you put it on the side chain, and suggests the keys that fit. Nothing changes until you pick one.
If the track isn't tuned to 440 Hz, Detect suggests its Reference A too, so notes are measured against the music they sit in.
2. Set how fast
- Retune speed is how fast a note is pulled into tune: 0 ms is the hard-tuned sound; tens of milliseconds sound natural.
- Note transition is how fast it moves to the next note, apart from within a note: short steps, long glides.
3. Leave the human parts alone
- Humanize goes slower on held notes, so their vibrato stays.
- Expression leaves slides, scoops and bends as sung: notes far from the target aren't touched.
- Amount says how much of the way to the note it goes.
- Formant keeps the voice sounding like itself when its pitch moves.
Natural or hard: the presets
Four presets set the main controls at once: Natural, Pop, Hard tune and Hard tune (keep vibrato). Start from one, then adjust by ear.
Record through it
Latency is 5 to 10 ms, depending on the Voice range you set, low enough for a singer to hear themselves tuned while recording. A narrower range is surer and needs less.
Write the melody
Hold MIDI notes and they become the target notes for as long as they're held: the way to make a vocal follow a melody exactly.