Use dither when a final export reduces the bit depth of processed audio to an integer format such as 16-bit PCM. The useful question is whether this export has fewer amplitude values available than the audio you have been processing. If you are only copying a file, there is no rounding operation for dither to address.
This guide uses Audacity’s dither documentation and an illustrative calculation. It does not report a listening test or claim that a particular noise difference will be audible on your equipment.
What is being rounded?
An integer PCM file has a finite set of sample values. Sixteen bits provide 65,536 possible values; twenty-four bits provide 16,777,216. The second count is 256 times the first. That follows from two raised to the bit depth, rather than from a measurement of a converter.
Those numbers describe the available amplitude values, not the number of samples taken each second. Reducing a file from 24-bit to 16-bit does not, by itself, change its sample rate. Our bitrate, sample-rate and bit-depth guide separates those three properties.
For a small calculation, suppose the only available values are spaced 0.1 apart. Your processing produces 0.24, but the file can store 0.2 or 0.3. Ordinary rounding chooses 0.2: subtract the intended 0.24 and the error is −0.04. For 0.26, it chooses 0.3, an error of +0.04. We chose that coarse grid to make the sums readable; a real 16-bit file has much smaller steps.
Across a whole waveform, those errors can follow a pattern. Audacity’s documentation identifies that relationship to the signal as the problem dither addresses. By adding controlled randomness during conversion, dither breaks up the regular rounding pattern. The cost is a small amount of noise. The resulting file still has the lower bit depth.
Three exports to distinguish
A WAV extension alone cannot tell you whether dither is needed. WAV can hold different sample formats. Look at the export’s actual encoding and bit depth, then compare them with the audio you have been processing.
For a processed high-precision project going to 16-bit PCM, dither belongs at that reduction. Keep the higher-precision project or master as well, so a later revision can start there. Reopening the 16-bit export and repeatedly saving new processed 16-bit versions introduces more conversions than necessary.
For an untouched 16-bit recording with only simple cuts or trims, the answer can differ. Audacity’s manual gives this as an exception: a same-depth uncompressed export with dither disabled can preserve the sample values. That exception depends on avoiding processing such as amplification, equalization or filtering. It is not a blanket instruction to disable dither whenever the original file says 16-bit.
The third case is a file copy. Select a WAV file in the file manager and duplicate it: the copy contains the same bytes. Nothing has recalculated its samples, so there is nowhere to apply dither. Renaming that copy makes no difference to its bit depth.
Check the setting that writes the file
Audacity distinguishes real-time conversion for playback from high-quality conversion used when writing audio to disk. Its Quality Preferences page explains that scope. Changing a playback dither option is not the same as choosing the conversion used for an export.
The manual documents several dither types, including triangular and shaped options. Their names describe different ways of distributing the added noise. Choosing one does not excuse an unnecessary conversion, and this source-based guide does not rank their sound through an unperformed comparison.
Before writing the deliverable, check its required encoding and bit depth. Keep your project or higher-precision master alongside it. If you later need to change the fade or level, open that master, make the edit, and export a new deliverable from there. That avoids starting the next round of processing with the already reduced version.