Yes: when you reduce a 24-bit or 32-bit float master to 16-bit, apply dither at the final bit-depth reduction. It trades signal-correlated rounding distortion for very low-level noise. An unchanged copy at the same or a higher bit depth needs no added dither, but processing can create higher-precision samples even when the input and output files are both 16-bit. Avoid unnecessary repeated reductions. WavSpec, the batch audio QC app from Automata Labs, offers bit-depth conversion with TPDF dither, re-measures delivered files, and reports their bit depth alongside the client’s other criteria.
This guide explains what dither does, which type to choose, and how to handle it in common tools.
Convert a batch to the delivery bit depth with TPDF dither, then prove it meets the spec.
Get it fromMicrosoft
What does dither actually do?
The Audacity manual’s page on dither explains why predictable rounding can create signal-related artifacts when samples are reduced to a coarser numerical grid. Dither introduces a small random component before quantization, making that error behave more like noise instead of following the signal.
Think of a quiet fade out. A 24-bit integer format has many more amplitude steps than a 16-bit format. As a signal gets very quiet, undithered rounding can produce audible distortion in a fade or reverb tail. Dither trades those correlated artifacts for a low noise floor. How noticeable either is depends on the signal and playback conditions.
How loud is dither noise?
It depends on the method and destination depth. The Audacity manual gives these approximate peak levels for its described export behavior:
| Export format | Dither behavior described by Audacity | Approximate peak level |
|---|---|---|
| 32-bit PCM WAV | No dither in the described export path | None |
| 24-bit PCM WAV | May be dithered | Around -130 dB relative to full scale |
| 16-bit PCM WAV | May be dithered | Around -80 dB relative to full scale |
These are implementation-specific examples, not universal TPDF noise levels or measurements of WavSpec. Peak noise and average noise floor are different quantities. Shaped dither can also have a higher peak while concentrating energy where it is less noticeable.
When should you dither, and when not?
The important reduction is from the working samples’ precision to the final PCM format, not necessarily from the original file’s bit-depth label to the export label.
- Dither: a 24-bit master to a 16-bit delivery, or a higher-precision floating-point mix reduced to 16-bit or 24-bit PCM.
- No added dither needed: an unchanged 16-bit file copied to 24-bit, a 24-bit file converted to 32-bit float, or an otherwise unchanged copy at the same depth.
- Dither at the final reduction: do gain, limiting, sample-rate conversion and other processing in the higher-precision working format first. Further processing followed by a fixed-depth export can require another quantization step.
For example, processing a 16-bit recording in 32-bit float and exporting to 16-bit is still a reduction from the processed samples. Audacity’s manual distinguishes that from simple cut, paste or trim edits that do not change sample values. Keeping a higher-precision working master avoids unnecessary repeated reductions; do not keep exporting and re-importing 16-bit intermediates.
Which type of dither should you choose?
Most tools offer some of the same choices.
| Type | What distinguishes it |
|---|---|
| Rectangular | A uniform probability distribution for the added noise |
| Triangular (TPDF) | A triangular probability distribution; a common unshaped choice |
| Shaped (noise shaping) | Redistributes noise toward frequencies where it is less noticeable |
Audacity’s manual describes triangular or shaped dither as useful subjective choices and lists shaped dither as its high-quality conversion default. Shaped dither can have a higher peak noise level even when it is less noticeable. Do not infer that triangular dither must have a lower noise level than rectangular merely from their names.
Whichever method the delivery workflow calls for, keep it consistent across the batch. WavSpec documents triangular (TPDF) dither for its bit-depth conversion.
Check the whole batch against one delivery spec and convert the files that need correction.
Get it fromMicrosoft
How do you dither in common tools?
Audacity. The manual locates the dither type under Edit, Preferences, Quality, in the High Quality Conversion settings. Keep the working track at higher precision, finish processing, choose the intended method, then export the final 16-bit PCM WAV. Menu labels can vary by version.
ffmpeg. The ffmpeg resampler options include dither_method, with values such as triangular, triangular_hp and noise-shaping methods such as shibata. The documented default is 0, so choose a method explicitly. This illustrative final-export command assumes in.wav is an already-processed 24-bit or 32-bit float master and out.wav is a new output file:
ffmpeg -i in.wav -af "aresample=osf=s16:dither_method=triangular" -c:a pcm_s16le out.wavIt requests triangular dither while reducing to 16-bit samples and writes a 16-bit PCM WAV. Do any level correction, limiting and needed sample-rate conversion in the higher-precision workflow before this final reduction. Do not treat this as a step to apply to every WAV regardless of its input precision.
DAWs. Many DAWs offer dither in the export or bounce dialog. Match the setting to the final reduction, and avoid enabling both a dither plug-in and the exporter’s dither for the same step. A 24-bit stem also needs consideration if it is being reduced from higher-precision processed samples.
What can go wrong in a delivery?
- No dither at all. A converter that simply truncates or rounds. The file meets the bit depth spec but quiet tails may sound grainy.
- Dither twice. A 16-bit file dithered, edited, then exported again with dither. Each pass adds noise.
- Gain after dither. Normalizing or limiting a 16-bit dithered file means rounding again. Do level correction first, then reduce bit depth.
- Mixed bit depths in one batch. Some files 24-bit and some 16-bit because they came from different sessions, so part of the batch fails the spec.
How does WavSpec handle bit depth in a batch?
WavSpec loads a batch of audio files, checks each one against a named delivery specification and corrects what can be corrected without touching the mix. For bit depth, that means bit-depth conversion with TPDF dither, and float-to-PCM conversion where a spec calls for PCM.
Its documented corrections also include gain adjustment, look-ahead true-peak limiting with 4x oversampled detection, and sample-rate conversion with a windowed-sinc resampler. A named client spec governs the batch, and corrected copies go to a separate delivery folder without overwriting originals. This description does not establish the app’s internal operation order or how many internal processing passes it uses.
WavSpec re-measures delivered files from disk and writes PDF, HTML, CSV and JSON reports with before and after values, pass or fail per criterion including sample rate, bit depth, channel layout and container, and each delivered file’s SHA-256 digest. The bit-depth entry verifies the delivered format; it does not by itself prove which dither was applied or how many times. Retain the relevant export settings as part of the workflow record. Processing is local and offline, with no account, cloud upload or telemetry.
Built-in specs include EBU R 128, ACX audiobook, Apple Podcasts and a generic 48 kHz 24-bit post delivery template, and you can create your own spec, for example a 44.1 kHz 16-bit delivery, and share it with a client as a file. Our audio delivery QC checklist covers the other checks to run before you send.
Frequently asked questions
Can you hear dither?
It is normally very quiet, but audibility depends on the method, bit depth, signal, playback level and equipment. Do not treat a general description of low-level noise as a guarantee that it is inaudible under every condition.
Should I dither when exporting MP3?
MP3 encoding is separate from PCM bit-depth reduction. Follow your encoder’s guidance. If an intermediate PCM reduction is actually required, handle dither at that reduction rather than assuming MP3 itself has a 16-bit delivery setting.
Do I need dither going from 32-bit float to 24-bit?
Reducing processed floating-point samples to 24-bit PCM can introduce quantization error, so the final conversion may use dither. Its level is much lower than at 16-bit; Audacity’s approximate export example is around -130 dB relative to full scale, not a universal audibility threshold.
Does WavSpec dither 24-bit deliveries?
WavSpec documents TPDF dither for bit-depth conversion, not an instruction to re-dither every unchanged 24-bit file. Set the required depth in the client spec and inspect the output report. The report confirms format criteria, not the signal’s complete processing history.
Try WavSpec
WavSpec is a one-time purchase with a free trial in the Microsoft Store; the listing shows the current price and system requirements. Run a 24-bit batch through a 16-bit spec and read the bit depth line in the report. More detail is on the WavSpec product page, and our guide to fixing true peak overs in WAV files covers the limiting step.
Use TPDF bit-depth conversion for delivery and inspect the exported format in the report.
Get it fromMicrosoft