Skip to main content

Music Production

Audio Quantization

Definition

Audio quantization maps a sample's amplitude to one of the values a digital format can represent. The difference between the original value and the chosen value is quantization error; this is different from moving notes to a rhythmic grid.

Sampling chooses when to measure a waveform. Quantization chooses how to represent each measured value with a limited set of numbers. In ordinary fixed-point PCM audio, bit depth sets the number of available amplitude codes. More bits provide a finer grid over the same full-scale range.

Round the level, not the beat

Imagine a simplified grid with steps of 0.1. Rounding an input of 0.26 to the nearest grid point gives 0.3, an error of +0.04 when defined as output minus input. An input of 0.24 becomes 0.2, an error of −0.04. These are teaching numbers, not the actual spacing of a 16-bit file. Neither operation moves the sample in time.

A 16-bit fixed-point sample has 65,536 possible codes; 24 bits provide 16,777,216. The extra codes improve potential resolution, but cannot by themselves eliminate microphone noise, analog distortion or converter limitations.

Error need not be random

Undithered rounding or truncation can produce error related to the signal, creating distortion rather than simply a steady background hiss. This becomes especially relevant to quiet passages and fades. Properly chosen dither adds a small random signal before quantization to reduce that signal-dependent behavior, at the cost of noise.

Raising the bit depth of an already quantized file does not recover its discarded precision. It can still be useful for subsequent processing, which may generate new values between the old grid points. Floating-point formats use a different numerical representation rather than one uniform amplitude grid over their entire range.

Two meanings of quantization

Rhythmic quantization adjusts note or event timing toward a beat grid. Audio amplitude quantization concerns the stored values of samples. A DAW can perform either operation; the shared word does not mean they fix the same problem.

Common questions

Does audio quantization tighten a performance's timing?

Not in this amplitude sense. It maps sample values to representable levels. Rhythmic quantization is the separate process that adjusts event timing.

Is quantization error always white noise?

No. Without suitable dither or other conditions, the error can depend on the signal and produce distortion. Treating it as independent white noise is a model with assumptions, not a universal fact.