.CHAPTER 3
.CHAPTER_TITLE "Phase Vocoding"
.HEADING 1 "Phase Vocoding"

.PP
Phase vocoding is a way of working with sound that uses the Fast Fourier Transform (FFT) to convert audio from the time domain to the frequency domain. This process decomposes a sound into composite sine waves with varying frequencies, amplitudes, and phases—these are called bins, and they represent the spectral content of the original sound.
.PP
FFT isn't the only method for spectral analysis, but it’s the most commonly used because it’s fast, efficient, and well-supported. Other methods, like wavelet transforms, offer compelling alternatives—not just because they can offer higher resolution, but because they analyze sound in fundamentally different ways. Where FFT assumes each frequency is present for the entire analysis window, wavelets allow for variable resolution across time and frequency, which can reveal more detail in transients, textures, and irregular events.
.PP
Wavelet transforms also hold deep creative potential. But their rewards come more slowly. Working with wavelets often means stepping into a compositional mindset where you're not producing sound immediately. You might be writing code, tuning parameters, and experimenting blindly for long stretches before anything interesting emerges. That’s not a drawback—it’s a different way of composing. We’ll explore wavelets later in the book, when we dig into more speculative and time-stretched approaches to synthesis. But that difficulty is part of what makes it powerful. New research means new sounds, and new sounds mean new culture.
.PP
That said, simply applying PV unit generators (UGens) to a single sample, or just modulating those UGens in isolation, rarely yields compelling results. To make phase vocoding expressive, you need either:
.LIST
.ITEM
Spectral morphing — combining the spectral data of two or more samples. This requires careful sample selection and matching of PV UGens, which we'll get into.
.ITEM
Decomposing the sample into multiple frequency or amplitude streams, processing them independently, and then resynthesizing them in parallel.
.LISTEND
.COLLATE