What Is Audio Dithering? Definition and Guide
Dithering is a small amount of noise added to audio when reducing bit depth. This sounds counterintuitive — adding noise to improve quality — but it works. Without dithering, reducing from 24-bit to 16-bit produces subtle distortion that is more unpleasant than the noise dithering introduces.
Why Bit Depth Reduction Causes Problems
Digital audio represents amplitude as numbers. At 24-bit, you have 16,777,216 possible amplitude values. At 16-bit (CD quality), you have 65,536. When you reduce bit depth, you must round each sample to the nearest available value. This rounding introduces quantization error — the difference between the original value and the rounded value. At normal listening levels, quantization error is inaudible. But in very quiet passages — the tail of a reverb, a faded ending, near silence — the error becomes a significant fraction of the signal. The result is quantization distortion: a harsh, buzzing artifact that is clearly artificial.
How Dithering Works
Dithering adds a tiny amount of random noise before quantization. This noise randomizes the rounding errors instead of letting them accumulate into patterns. Random noise is perceptually less objectionable than patterned quantization distortion — the ear accepts noise as natural, while it rejects the repeating artifacts of undithered quantization. The noise is typically added at the threshold of audibility — around -90 to -96 dBFS for 16-bit audio. Shaped dithering (like TPDF or noise shaping curves) pushes the dither noise into frequencies where hearing is least sensitive, usually above 10 kHz, making it even less perceptible.
Types of Dithering
RPDF (Rectangular Probability Density Function): Simple random noise. Reduces quantization distortion but adds flat-spectrum noise. TPDF (Triangular Probability Density Function): Two layers of RPDF. Better error decorrelation. The standard for most mastering applications. Noise-shaped dithering: Applies a filter to concentrate dither noise in less audible frequency ranges. Common implementations include POW-r (three levels of noise shaping) and UV22 from Apogee. Noise shaping can reduce perceived noise by 6-10 dB compared to flat-spectrum TPDF. Most DAW dithering options include TPDF and at least one noise-shaped option.
When to Apply Dithering
Apply dithering exactly once: at the final export step when reducing bit depth. Specifically: when going from 24-bit or 32-bit float internal processing to 16-bit for CD or distribution. Never dither in the middle of a session. Never dither when exporting to 24-bit — there is no bit depth reduction happening. Never dither multiple times — it adds noise unnecessarily. In your DAW, enable dithering in the export or bounce dialog. In Pro Tools, select Pow-r or UV22HR. In Logic Pro, select the dither setting in the Export dialog. In Ableton Live, use the Dither setting in Export. AudioUtils applies appropriate dithering automatically when the output format requires a bit depth reduction.
Dithering in Practical Audio Work
Most listeners cannot consciously identify undithered quantization noise in real-world listening conditions. But mastering engineers and experienced listeners notice it in A/B comparisons, especially on classical music, acoustic recordings, and anything with long, quiet decay tails. For casual conversions — voice memos, podcast exports, video game audio — the difference is academic. For commercial music releases, vinyl transfers, and archival masters destined for CD, always apply dithering at the 16-bit export stage. When converting audio files with AudioUtils for distribution, dithering is handled transparently. You do not need to think about it unless you are mastering for CD.