Autotune Settings for Rap: Exact Numbers by Style

Set the key and scale first, then start here. Hard-tune and ad-libs want retune speed near 0 ms with correction amount at 100%. Melodic rap sits around 10 to 20 ms at 100%, and transparent cleanup lives near 40 to 80 ms at 70 to 90%. Retune speed decides how the tuning sounds; amount decides how much of it you hear.

Best autotune settings for rap (start here)

Copy these, then change one control at a time. Retune speed is sometimes labeled Speed or Correction Speed; amount may be Mix or Strength.

Disclosure: we build audio plugins at Ghostnote, and one is a pitch corrector. It appears in one labeled section near the end. Everything else here works in the tuner you already own. No affiliate links.

Step 0: what key is the beat in?

The most common reason a tuned rap vocal sounds broken: the plugin is set to chromatic or the wrong key, so it drags every note to the nearest semitone instead of the right one. The tuner cannot tell, and a confidently wrong note does more damage than an untouched one.

Three ways to get the key:

1. Ask whoever made the beat. On lease sites the key is usually in the file name or listing.
2. Play a root note against the loop and walk up until it stops fighting the 808.
3. Use a key-detection utility, then confirm by ear. Detection reads the loudest harmonic content, so a heavily sampled hook can report the relative minor rather than the tonic.

Then narrow the scale. A hook rarely uses seven notes. If the melody moves between three, set the scale to those three. That removes most of the warble people blame on the plugin. If takes still come out mechanical, work through the seven usual causes of robotic autotune.

What does retune speed do?

A tuner runs three steps hundreds of times a second: estimate the fundamental frequency, pick a target from the key and scale, then move the measured pitch toward it. Retune speed controls the third step and nothing else.

At 0 ms, correction is instant, so onset scoops, vibrato and drift inside a note are all flattened. That is hard-tune. At 200 ms, it arrives slowly enough that vibrato and slides pass through, and only the average pitch gets nudged. Style control, not quality control.

Humanize, flex-tune and natural vibrato

Three other controls hand back some of what retune speed takes. Antares Auto-Tune uses the names Humanize, Flex-Tune and Natural Vibrato; other vendors label the same jobs differently. Set retune speed first, then these.

• Humanize corrects sustained notes more slowly than onsets, so held vowels keep moving while attacks stay locked. It saves melodic hooks.
• Flex-tune leaves notes already near the target alone and pulls only the ones that miss. Off for hard-tune, useful on spoken bars.
• Natural vibrato scales the vibrato already in the take. Leave it neutral and skip added vibrato: on a rapped bar it reads as a plugin working.

The part that separates good tuners from bad ones

Retune speed is a smoothing control, and there are two ways to build it. A plugin can smooth the target pitch, or it can smooth the correction it applies. Smooth the target and a deliberate note change becomes a slide, because the target has to travel between notes. That glide is tuner portamento. Smooth the correction instead and the time constant lands on the offset between voice and target. That offset barely moves when you jump a whole tone, so a note change passes straight through, while drift inside a held note still gets flattened.

That is how we built our SIREN. Its correction time constant runs from roughly 2 ms at retune 0% to about 500 ms at 100%. In our probe suite, a note held 30 cents flat at A3 lands within 5 cents of target, with under 2 dB of level shift. An A3 to B3 jump settles 2 cents from B3 within 0.3 s, and in-tune material passes unchanged.

Detection is a YIN estimator with CMNDF and parabolic refinement, 60 to 1100 Hz at about a 5 ms hop, under 3 cents of error on steady tones. Breaths, consonants and silence are rejected as unpitched. YIN comes from de Cheveigné and Kawahara's 2002 JASA paper. The correction is time-domain PSOLA, Hann grains at twice the detected period, published by Moulines and Charpentier in 1990.

Autotune settings by style, explained

Hard-tune (the locked, stepped sound)

Retune 0, amount 100%, scale trimmed to the notes of the hook. The stepped quality comes from instant correction, not from a high amount, which is why pulling amount down gives a blurry double image of the voice rather than a softer effect. Two habits beat the numbers. Sing close to the target, because a note more than halfway to its neighbor steps somewhere you did not want. Then print the tuned vocal before you mix it.

Melodic rap and sung hooks

Retune 10 to 20 ms, amount 100%, a little humanize. Most current rap vocals live here: fast enough that the pitch reads locked, slow enough that a phrase tail can bend.

Transparent cleanup on straight bars

Retune 40 to 80 ms, amount 70 to 90%. Spoken rap moves in pitch by design, and the emphasis in a punchline is often a deliberate rise, so heavy correction flattens the delivery. Leave the scale on the key. Chromatic is defensible on a spoken verse, where the tuner only keeps long vowels from sagging.

Ad-libs, doubles and background vocals

Retune 0 to 10 ms, amount 100%, same key and scale as the lead. Ad-libs carry the effect in modern rap, so tuning them harder than the lead is normal. Tune each lane separately, not a group bus: grain-based correction tracks one voice at a time, and a stacked bus hands it two fundamentals to argue about.

What is formant preservation, and when should you turn it off?

Your voice has two parts. The vocal folds set pitch. The throat, mouth and nose filter that signal into resonant peaks called formants, which carry the vowel and the perceived size of the speaker. That is the source-filter model Gunnar Fant described in 1960.

Move pitch by resampling and the formants move with it: a voice pitched up sounds smaller, one pitched down sounds like a cartoon villain. Formant preservation holds the peaks in place while the fundamental moves, so a corrected note still sounds like the same body. In our SIREN it is on by default, and the correction runs one shared grain schedule across both channels so a stereo part stays coherent. Switch it off and grains get resampled: chipmunk or demon, on purpose.

Leave it on for anything that should sound like a person. If a corrected note sounds thinner than its neighbors, check this switch before reaching for EQ.

Where does autotune go in the vocal chain?

After corrective EQ, de-essing and level control. Before tonal EQ, reverb and delay. The reasoning is mechanical, not aesthetic:

• After a high-pass and subtractive EQ. Rumble and proximity buildup give the detector low-frequency energy to misread.
• After de-essing. Sibilance is noise, not pitch. A quieter ess is one less thing for the detector to reject.
• After level control. A leveled take gives the detector a steadier signal, so note capture stays consistent across the whole verse.
• Before time-based effects. A reverberant or delayed signal smears the pitch the tuner is measuring.

The stage-by-stage version, with settings for every box, is in our rap vocal chain guide. One exception: when hard-tune is the sound of the record, put the tuner first and print it, so everything after mixes a tuned instrument instead of fighting a repair.

Watch latency while tracking. Grain-based correction needs lookahead: our SIREN reports 2000 samples of it to the host, about 42 ms at 48 kHz. Most hosts compensate reported plugin latency on playback, per Steinberg's VST 3 developer documentation, but an artist hearing themselves live still needs a low-latency monitor path.

How do you record a take that tunes cleanly?

Tuning is a finishing process, not a rescue. Everything below is cheaper than fixing pitch later.

• Print a guide the artist can hear. Two bars of the root under the hook, quiet in the phones.
• Stay in a supported part of the range. A note at the top of someone's chest voice arrives breathy, and correcting it gives you a tuned breathy note. Drop the hook a whole tone and recut.
• Punch on note boundaries. Splicing mid-vowel leaves a pitch discontinuity the tuner reads as a jump.
• One voice per lane. Overlapping ad-libs on one track is where detection falls apart.
• Do not tune twice. A tuner on the take plus another on the bus stacks two sets of grain artifacts.

Which autotune plugin should you use?

Three of these run in realtime, one does not. Antares makes Auto-Tune, which named the sound and remains the reference. Waves sells a tuner popular for tracking. Auburn Sounds offers Graillon in a free edition, a strong start for hard-tune work. Celemony's Melodyne edits notes offline: better for surgical work, wrong for a sound you perform through. We compared the approaches in Melodyne vs autotune. Prices change often.

Check what you already own first. Logic Pro includes a Pitch Correction plugin whose Response control is retune speed under another name, plus Flex Pitch for offline note editing (Apple's Logic Pro User Guide). Image-Line's Pitcher does realtime correction inside FL Studio (Image-Line). Control names and ranges differ, so match the behavior above, not the number.

From Our Own Line

We make SIREN, a realtime corrector with a key-aware note grid. In-key lanes glow, off-scale notes show as red blocks with cent labels, and each captured note is a block you drag. Capture triggers at more than 0.6 semitone from the running median, or after an unvoiced gap over 80 ms, with a 60 ms minimum and room for 4000 notes. A hook full of red blocks tells you the scale is wrong in about a second.

Its four presets map onto the styles above. Retune 0% is the 2 ms end of the time constant.

What it is not: a polyphonic editor, or a substitute for recutting a weak take. It corrects one voice at a time, the pitch-shift ratio is clamped between 0.5× and 2×, and bypass is bit-exact, so an A/B compares the correction and nothing else. It is $39, VST3 on macOS and Windows plus AU on macOS, keyless. No fake sales. No intro-price games. The price is the price. The rest of the line, including a free de-esser, is on the Ghostnote plugins page.

Frequently Asked Questions

What retune speed should I use for rap vocals?

Start at 10 to 20 ms for melodic rap and 0 ms for hard-tune. Bars that only drift want 40 to 80 ms. Retune speed sets how fast the plugin moves your pitch to the target, so fast values flatten scoops and vibrato.

What are the best autotune settings for a hard-tune sound?

Retune speed 0, amount 100%, humanize off, formant preservation on, and a scale trimmed to the notes the hook uses. The stepped quality comes from instant correction, not a high amount. If it lands on odd notes, the scale is wrong.

What does the humanize control do?

Humanize applies a slower correction to sustained notes than to onsets, so a held vowel keeps its natural movement while the attack of each word still lands locked. Turn it off for hard-tune. A little of it is what keeps a melodic hook from sounding stiff.

Should autotune go before or after compression?

After. Put the tuner behind corrective EQ, de-essing and level control, and ahead of tonal EQ, reverb and delay. A leveled, de-essed signal gives the detector a steadier read, and time-based effects smear the pitch it measures; hard-tune printed as an effect is the exception.

Can autotune fix an out-of-tune take?

It fixes pitch, not phrasing, support or timing. Corrections under a semitone sound clean; past that the plugin moves grains far enough that you hear the work, so recut the line. Tuning a breathy, unsupported note gives you a tuned breathy note.

Do I need to know the key of the beat to use autotune?

In almost every case, yes. Chromatic tuning snaps to the nearest semitone, often a note that is not in the song, which is the most common reason tuned rap sounds warbly. Set the key, or restrict the scale to the notes the hook uses.