Vocal editing: comping, timing, tuning, Melodyne and Auto-Tune.

After vocal recording, there is almost never just one take: there are five, eight or twelve, and none gets everything right. Vocal editing turns that material into a single line sounding like a continuous performance, without revealing the work.

The process involves three decisions and one warning. First, comping: which parts of each take to keep. Second, timing: how far to move the phrase without killing the groove. Third, tuning: how much to correct and which tool to use, Melodyne or Auto-Tune. The warning is that heavy tuning is a deliberate stylistic choice rather than a technical fault.

From several takes to one voice

Comping means creating a composite take: select the best moments from several performances of the same passage and join them into a track sounding like one performance. The goal is the most convincing combination, rather than the cleanest take: the phrase from take two, the turn from seven, the breath that works from four.

This matters in DAWs with dedicated tools and DJ software with basic editing, because the problem is always the same: no take has the best entry, pitch and ending together. A good comp is invisible; a poor one reveals itself through cuts sounding like position changes, interrupted breaths or an abrupt timbre change midway through a word.

Comping: building the best version take by take

Each DAW handles the same workflow differently. Logic Pro groups overlapping takes into folders and uses Quick Swipe Comping: drag over the desired section and the folder switches to that take. You can change each section’s start and end, move sections, and save named comps to return to earlier versions. Ableton Live arranges takes in parallel take lanes below the main track: audition each lane, select desired material and copy it to the main track, where its source is highlighted in the track colour.

The method is identical in both: listen to the complete section, choose the longest passage possible from one take, and cut only where the phrase permits. Cuts halfway through a syllable or directly on a consonant expose the assembly; cuts between words or in natural silences do not. Work in blocks of bars rather than isolated notes: every extra cut creates another chance for pitch or volume to change between takes.

Timing editing: how far to move and when to stop

Vocal timing is corrected by section rather than globally. Move what the ear hears as a mistake: entries late against the kick, words swallowing a bar’s first beat, or a verse’s final syllable ending early before the next section.

Do not move what belongs to the performance: intentional dragging that creates swing, slightly laid-back phrases or the singer’s own breathing rhythm. Gridding the entire vocal makes the line perfect and lifeless simultaneously, because the ear hears absent microvariation as programming.

A practical rule: adjust with the track playing, then solo, then return to the track. If it only works solo, you have moved too far. Heavy quantisation makes sense for short fragments needing exact placement and styles where rigid precision is part of the sound; elsewhere, correct only what is necessary.

Natural tuning versus a stylised effect

Tuning has two different goals requiring different settings. Natural correction keeps notes in tune without audible processing: smooth movement, respect for vibrato and correction below the obvious threshold. The stylised effect seeks the opposite: notes jumping instantly to the centre with the robotic character that made a particular plugin famous.

Antares documents this decision in Retune Speed: measured in milliseconds, 0 produces the iconic Auto-Tune effect with instant pitch changes, while values between 10 and 50 are typical for more natural correction because they allow vibrato and expressive gestures through. Humanize applies slower correction only to the long portion of sustained notes, while Flex Tune lets the voice deviate within a window around the note before correction.

Choose the correct scale before adjusting anything: tuning against the wrong scale turns correction into an error. You also need the right tool for the material: monophonic versus polyphonic.

Melodyne and Auto-Tune: monophony and polyphony

The tools start from different philosophies. Auto-Tune works in real time on a monophonic signal: one voice or an instrument playing one note at a time. Antares explicitly states that it cannot correctly tune several voices or instruments recorded on the same track. Its workflow uses a few parameters: key and scale, retune speed, Humanize and, live, latency.

Melodyne works with a file: it analyses audio, detects notes and shows blobs that can be moved in pitch and time. Its Melodic algorithm is intended for vocals because they are monophonic, and integrates sibilant detection so “s” sounds do not move with the note. Polyphonic algorithms exist for instruments playing several simultaneous notes. Celemony warns that changing algorithm after editing forces redetection and discards edits, so choose before starting.

For vocals, the boundary is simple: with one voice on a track, either works; with doubled harmonies or a choir on one channel, neither separates them well. Work on separate vocal tracks in that case.

Formants and the robotic voice

Formants are resonances giving the vocal tract its timbre: a low and a high voice can sing the same note yet sound like different people. Raising a note in software without doing anything else shifts the entire spectrum, including those resonances, producing a toy-like voice.

Formant management distinguishes transparent correction from an obvious effect. Correction plugins include formant controls to preserve timbre while pitch changes, and Antares itself provides linked formant and transposition controls in its more capable versions. The listening rule is simple: if raising a vocal by three semitones makes it sound like a caricature, the problem is the formant rather than the amount of tuning.

Harmonizers and harmonic doubles

Harmonizers generate derived voices that sound like backing vocals: they take the lead and create detuned or harmonically shifted doubles to thicken a phrase. Production uses them in two ways: as an ideal double reinforcing choruses, or as an obvious effect when the harmony is intentionally audible.

The criterion concerns the mix rather than the plugin. A double very similar to the lead causes comb filtering and blur; a shifted double with a different character functions as a second voice. A double always competes with the lead for the same space, so its level and EQ are decided later during vocal mixing, where doubles and ad-libs support without covering it.

Next step

With the take assembled, synchronised and tuned, the vocal is ready for the chain: vocal mixing with EQ, compression, de-essing, saturation, reverb and delay.