Dither
A trace of noise before bit-depth reduction turns quantization distortion into a steady hiss.
Related
The question
Reduce bit depth, or print a floating-point mix to a fixed-point file, and the quietest parts hit the steps.
How it works
Quantization rounds values to the nearest step, and on quiet, periodic material the error lines up into audible grain. Dither adds random noise of about one least significant bit before rounding, so the error decorrelates and the floor rises slightly. Noise shaping then pushes it toward high frequencies, where the ear is less sensitive.
Settings and practice
Use TPDF dither with moderate shaping for 16-bit delivery. Add it once at the final reduction, then change no gain and encode nothing further. At 24 bit it is usually unnecessary, since the floor sits below the recording. Choose lighter shaping for bright material.
What to listen for
In quiet lute pieces, listen for the steps. At a song's end, hear whether voice and piano turn sandy as they fade. In a long nocturne tail, check whether the hiss stays even. Without dither the tail dries in patches; with it you hear continuous fine sand.
Common misconceptions
Dither adds noise, so it must reduce quality; what it trades is distortion for noise, which wears better. It is not needed on every export, only when reducing depth. Adding it at 24 bit achieves nothing. It cannot repair already clipped material.
Listen in the library
All tracks from midicn-lib — inline badge is the license tier (C1 commercial / C2 non-commercial / C3 study only).
Sources
- Lipschitz 与 Vanderkooy 关于抖动的公开论文,依原文核查,未抄文本
- Pohlmann, Principles of Digital Audio, quantization chapters — background checked, not quoted
Used for fact-checking only; all prose is original.