Quantization is the second half of digitizing audio, after sampling: each measured sample, a continuous voltage, is rounded to the nearest value the bit depth can represent. With 16 bits there are 65,536 possible levels; with 24 bits, over 16 million. The rounding error is quantization noise, and at low bit depths it correlates with the signal, producing audible distortion on quiet passages rather than benign hiss. Dithering — adding a tiny controlled noise before rounding — decorrelates that error and trades the distortion for a constant, unobtrusive noise floor. Quantization is why bit depth sets dynamic range, and why you dither whenever you reduce it.
