Dither in Mastering
Add dither only at the final bit-depth reduction, and account for noise shaping and peak margin on 16-bit releases.
Related
A 16-bit master fails first in its quietest seconds. Dither is not a noise layer over the whole record; it is the choice made at the final truncation, trading signal-linked quantization distortion for steady, low-level hiss.
When dither belongs
Only at the last bit-depth reduction. Mixing, mastering and limiting happen in float, and a 24-bit export usually needs no dither because its noise floor already sits below most recordings. The real case is 16-bit delivery: CD, some broadcast chains and older players. If the same version will be encoded lossily, dither to 16 bit first and encode afterwards; never add dither after the codec.
How the error becomes audible
Quantization snaps continuous amplitude to the nearest step. On soft notes, long tones and periodic material the error lines up into a pattern and sounds gritty. Dither adds about one least significant bit of random noise before truncation so the error decorrelates from the signal, at the cost of a slightly higher floor. Noise shaping pushes that error toward high frequencies where hearing is less sensitive, buying a quieter midrange. Heavier shaping means more high-frequency noise and more trouble in low-bitrate codecs and lacquer cutting.
Settings and delivery
Bit depth decides whether dither is needed, the shaping curve decides where the noise goes, and true-peak margin decides whether encoding clips.
| Delivery | Bit depth | Dither | Shaping | Peak margin |
|---|---|---|---|---|
| CD / 16-bit WAV | 16 bit | TPDF | Light to medium | -1 dBTP |
| Streaming master | 24 bit | Usually none | None | -1 dBTP |
| Archive master | 24 or 32-bit float | None | None | 3 to 6 dB left |
| Lossy preview | 16-bit source | Once | Light | Margin before encoding |
What to listen for
Hear the tail of a soft note, piano decay and reverb. The difference is not in the loud part but in the last second or two: steady hiss or pitch-linked grit. The piano figure below starts softly and leaves a long tail.
Where it goes wrong
Dither after lossy encoding only adds noise to an already damaged signal. A 16-bit dither on a 24-bit master raises the floor for nothing. Heavy shaping can turn into high-frequency hiss in low-bitrate MP3 or on vinyl. If the distributor asks for an undithered master, send that and attach a 16-bit reference. Dither cannot fix overload, clipping or a bad recording; it only handles the final quantization.
Listen in the library
All tracks from midicn-lib — inline badge is the license tier (C1 commercial / C2 non-commercial / C3 study only).
Along the same line
Sources
- Pohlmann, Principles of Digital Audio（McGraw-Hill）— 量化、抖动与噪声整形章节。
- Lipschitz 与 Vanderkooy 在 AES 会议上关于量化与抖动噪声的论文 — 误差去相关与噪声整形的依据。
- ITU-R BS.1770 与 EBU R128 — 真峰值与交付余量的计量口径。