Pitch and duration are locked together when a file is played faster. Separating them means measuring what the signal is made of, moving each part of it, and putting the result back on the original timeline.
Semitones and fine tune
Semitones moves the pitch in whole steps, up or down by 12. Fine tune adds up to 50 cents either way, which is half a semitone.
At exactly 0 semitones and 0 cents the wet path passes the input through unchanged, and the crossover happens over 20 ms, so switching a shift in or out during playback does not step.
Common intervals: 12 is an octave, 7 a fifth, 5 a fourth, and 3 or 4 the minor and major third.
Mix
Mix blends the shifted signal against the original. At 100% only the shifted version is heard.
Lower settings are where the interesting uses are. A 5 cent detune at 40% mix thickens a vocal without an audible pitch change, and a 12 semitone shift at 25% adds an octave under a bass line while the original note stays in front.
How each partial is moved
The file is cut into overlapping frames of about 45 ms and transformed into a spectrum. A note does not land on one frequency bin, it lands on a peak with a small cluster of bins around it, and how fast the phase of that peak turns from one frame to the next says exactly which frequency it is sitting on, well past the resolution of the bins themselves.
Each peak and its cluster are then slid together to the new frequency, and the phase they are resynthesised at advances by the shifted frequency every frame and is never reset. Continuity is the point. Anything that restarts a phase mid-note adds a small rotation to it, and a rotation repeated a few hundred times a second is a frequency of its own sitting on top of the interval you asked for.
Where it struggles
Percussion softens. A frame 45 ms long cannot place a snare hit more precisely than that, so the attack spreads over the width of the frame. Sustained material, single instruments, and voices come through better than a dense mix.
Broadband content thins out on large upward shifts. Cymbals, breath, and distortion are noise rather than partials, and the top of that noise moves past the sample rate’s own ceiling and is dropped instead of folding back down as false tones.
Shifts of more than about 5 semitones are audible as processing. For a clean transposition of a musical part, a smaller shift on a stem sounds better than a large one on a finished mix.
Example: a backing vocal double
Semitones 0, fine tune 12 cents, mix 45%.
The take widens and sits behind the lead without landing on a different note.