Pitch shifting is the process of raising or lowering the perceived fundamental frequency of an audio signal, and it represents one of the more technically challenging transformations in digital audio processing. To understand why, it helps to grasp what pitch actually is: pitch is the perceptual attribute of sound that allows us to order sounds on a frequency-related scale from low to high. It corresponds primarily to fundamental frequency—the lowest frequency component of a harmonic series—but is also influenced by the relative amplitudes and phases of overtones, the spectral envelope (formant structure), and even the duration and loudness of the sound.

The simplest method of pitch shifting is resampling: to raise pitch, you play back fewer samples per unit time (effectively speeding up the audio), which shortens the duration proportionally. Lowering pitch means inserting interpolated samples (slowing down playback). This approach is computationally trivial but inherently couples pitch and duration changes—a limitation that is unacceptable for most practical applications where you need to change pitch while preserving the original timing.

Modern pitch-shifting algorithms solve this problem through two main approaches. The PSOLA (Pitch Synchronous Overlap and Add) family of algorithms works in the time domain and is particularly effective for monophonic signals like solo voice or single instruments. PSOLA first identifies the pitch period of the signal by detecting the fundamental frequency at each point in time. It then segments the audio into pitch-synchronous windows centered on each pitch period, repositions these windows at intervals corresponding to the desired new pitch period, and overlaps-and-adds them to reconstruct the signal. For pitch increases, windows are placed closer together; for decreases, farther apart. A complementary time-stretching step compensates for the duration change, resulting in pitch modification with preserved duration.

The phase vocoder approach operates in the frequency domain by decomposing audio into its spectral components using the Short-Time Fourier Transform, then scaling all frequency bins by the desired pitch ratio, and resynthesizing via inverse FFT. This approach handles polyphonic audio (chords, full mixes) better than PSOLA since it does not require pitch detection, but it can introduce phase coherence artifacts that manifest as a metallic or chorusing quality, particularly with large pitch shifts.

An important nuance in pitch shifting is the distinction between pitch transposition and formant preservation. When you naturally sing a higher note, your vocal tract shape stays roughly the same, meaning the formant frequencies—the resonant peaks that give your voice its characteristic vowel quality—remain constant while the fundamental frequency rises. Simple pitch shifting raises both fundamental and formants together, which is why extreme pitch shifts make voices sound unnatural (the chipmunk effect). Advanced pitch shifters can independently control formants and fundamental frequency, producing more natural-sounding results for vocal transposition.