SamplingMusic theory

Pitch A Vocal Loop Into Your Key Without The Chipmunk

Pitch shifting moves formants along with the notes, and the ear reads formants as body size. Why that sounds wrong, and what to do instead.

All articles

Moving A Voice Costs Something

Pitching a sample is a different operation from playing it faster, and neither one is free. Once you know the loop's root, whether from finding the key by ear or from the file name, the real question is whether to move it at all. A pitch shift multiplies the whole spectrum of the voice, including the parts that tell a listener how big the singer is. That is where the chipmunk comes from. It is not a quality setting you can turn up, and it is not something a better algorithm removes.

Formants Are Body Size

A voice carries two independent pieces of information. There is the pitch of the vocal folds, and there are the resonances of the throat and mouth sitting on top of it. Those resonances are the formants, and their frequencies depend on the size of the vocal tract rather than on the note being sung. A large person singing high still has low formants, which is exactly how you can tell a child from an adult singing the same note.

A plain pitch shift multiplies every frequency in the file by the same ratio, so the formants travel with the note. A couple of semitones up and the listener hears a slightly smaller person, which usually passes. Push further and the vocal tract being described stops corresponding to any real body, so the voice stops reading as a singer and starts reading as a processed sample. The note is correct and the source is no longer believable.

Formant Preservation Is A Second Process

Formant preserving modes shift the pitch and then push the resonances back toward where they started. That is genuinely different work: the algorithm has to decide which part of the spectrum is the body of the voice and hold it still while everything else moves. When it decides well you get a wider usable range. When it decides badly you get a hollow or phasey vowel, and no EQ afterwards puts the missing formant back.

The practical range stays smaller than the interface implies. Two or three semitones in either direction usually survives on a sung line, and artefacts climb quickly past that, faster on long vowels than on short adlibs. Audition the longest held note rather than the first word, because that is where smearing shows. Picking a phrase already near your key is cheaper than any algorithm, and a pack like MIRA spreads its Turkish vocals over four categories, so there is usually another take to try first.

Four Ways Around The Shift

Move The Track To The Loop

The cheapest correction is the one that costs the vocal nothing. Transpose your MIDI, retune the 808, move the chords, and leave the sample where it was recorded. A key is an arbitrary starting point in an unfinished beat, while a recorded voice is the one element in the project that cannot be rendered again at a different pitch for free.

Use The Loop An Octave Away

An octave is the one interval simple enough that the whole spectrum stays in a musical relationship, because doubling every frequency lands on the same note name. Formants still move, so a full octave up will sound obviously processed. An octave down often reads as a deliberate effect instead of a mistake, which is why pitched down vocal textures are common and pitched up ones are not.

Correct One Slice, Not The Loop

Usually only a couple of notes actually clash. Chop the phrase first, shift the offending slices alone, and leave the rest of the file at its original formants. That keeps most of the take untouched and confines the artefacts to a note or two. The cutting has its own rules, covered in chopping a vocal loop so it stays in time.

Keep The Vowel, Drop The Rest

Most of a vocal's melody is carried on the sustained vowels. If a phrase refuses to fit your key, keep the vowel that does fit, cut the words around it, and repeat it as a texture. One held vowel in the right key does more for a beat than a whole line arguing with the chords underneath it.

Signs You Have Pitched Too Far

  • The singer sounds younger or smaller than on the original file, because the formants moved with the note.
  • Long notes warble while short words survive, because stretching artefacts accumulate across the length of a held vowel.
  • The vowel sounds hollow or nasal, because formant preservation is holding the wrong region of the spectrum still.
  • Consonants smear into the note that follows, because the algorithm is treating a noise burst as pitched material.
  • The vocal suddenly needs more EQ than it did, because the shift moved the body of the voice out of the range you had already carved for it.
  • The loop sat better in the mix before it was in the right key, because the untouched file carried detail the shift destroyed.

Turkish Vocals, Volume One

MIRA collects Turkish vocals across four categories, adlibs and phrases through to stem by stem construction kits, cleared for release with no royalties.

See the pack

Guides

You may also like

Our latest news

More guides