Reducing bit depth adds harmonic quantization noise; dithering trades it for benign broadband noise

Digital audio amplitude is stored as integers with n bits, giving 2^n quantization levels; each sample is rounded to the nearest level. When too few bits are used — or at very low signal levels where only a few bits of resolution are active — the rounding error becomes correlated with the signal and is heard as harsh, harmonically-related distortion (quantization noise) rather than random hiss. Dithering adds a small amount of random noise, scaled to about one quantization level, before quantization/ADC conversion. This randomises the rounding error, converting the structured harmonic distortion into wideband noise, which the ear tolerates far better. Dither is especially important when reducing bit depth (e.g. 24-bit to 16-bit for CD), where the discarded lower bits would otherwise distort quiet passages and reverb tails; mastering engineers apply triangular-probability-density dither before the reduction. A subsequent low-pass filter can reduce the added noise. Lo-fi aesthetics deliberately exploit undithered bit-reduction artefacts.

Examples

8-bit audio (256 levels) at low amplitude has clearly audible quantization distortion; adding one level of dither noise before 4-bit reduction keeps drums recognisable despite severe degradation. Mastering 24→16-bit: apply TPDF dither before the bit-depth reduction to preserve reverb tails. In Max, degrade~ reduces bit depth and sample rate; feed it noise~ scaled to 1/quantization_levels to simulate dithering.

Assessment

Explain why a soft signal recorded at 4- or 8-bit without dither sounds harsh rather than merely quiet. Describe the audible difference between a 4-bit undithered and a 4-bit dithered signal of the same material, and explain why dithering is described as trading distortion for noise and what it does to the distortion spectrum.