An engineer’s ability to take an “off note” and bump it to a correct note is a well known fact. Made most popular by the use of Antares Auto-Tune by artists such as Daft Punk, Cher, and T-Pain, Auto-Tune has become a commonly understood (sometimes mis-understood) concept and has even shown up as an iPhone app.
Tuning software has been around for well over a decade, and has been used in records in far more transparent ways for awhile now. In fact, it’s been a lifesaver to many a recording. Sometimes you get that magical take (or only one take) that just has that slightly off moment. Tuning gives the engineer the choice to leave the take natural, or make it “perfect.”
Pitch Correction vs. Pitch Shifting
Pitch Correction is not simply finding the intended pitch and gluing the off note to it. That’s called Pitch Shifting. While Shifting and Correction have a lot of things in common – there is one fundamental difference. Instruments, particularly the human voice, have harmonic signatures.
These signatures, in conjunction with overtone patterns, allow us to identify when an instrument is a guitar, a flute, a sine wave, or a voice. The voice in particular is manually shaped with various harmonic signatures, called Formants, that yield vowel sounds.
Our natural overtone pattern, and the resonance shaping we make with our glottis, mouth shape, and nasal cavity come together to form our unique sound. When we sing higher or lower notes, some characteristics change, but our Formants (our harmonic signature) actually stays the same.
After all, just because one note is lower than the other doesn’t mean our mouth or nose changes dimension in order to produce the lower note.
Pitch Correction accounts for this harmonic signature, these formants, and rearranges our sound to the desired pitch while preserving the harmonic shape.
How It Works
This part is pretty complicated, and the most technical bits are still outside of my understanding. However, I’ll lay down the basic info. Pitch Correction uses a variant of phase based vocoding. So, when people call Pitch Correction software “the vocoder” – they aren’t totally incorrect. Pitch Correction is essentially a very specific vocoder.
Phase Vocoding sounds complicated. It is and it isn’t. At the most basic level a vocoder isn’t terribly complex. It literally “codes voice.” Our voice has different amplitude shapes at different frequencies as we pronounce words (particularly with vowel sounds). The vocoder reads the incoming volume levels at different frequency bands and figures out the shapes.
These shapes control a set of frequency filters in the vocoder which are then applied to a different signal. While we most commonly think of this as being done by a compuer, it can actually be accomplished in the analog world relatively simply. But, the more frequency bands and filters used, the more accurate the vocoding will be – and computers allow for extremely high numbers of bands and filters to work simultaneously.
Here’s the complicated part. In traditional vocoding, the voice may act as the modifying control for a simple wave shape or sound. In Pitch Correction, the voice essentially acts as the controller for a Pitch Shifted version of itself.
Not only that, but the pitch shifted version is kept at the same or a similar time boundary as the original. This is done through mathematical algorithms that involve Fourier transformation and re-synthesis to get down to the basic structure of the sound, and re-create it at a different pitch without changing the rhythm – or making “intelligent” time changes that keep it as close as possible with minimal artifacts.
The bottom line: Pitch Correction changes the pitch, but keeps the harmonic signature of the original (or as close to it as possible).