Understanding Dithering

Dithering is a technique rooted in the earliest days of digital audio, designed to combat a fundamental problem: quantization error. When an analog signal is converted to digital, it is sampled and each sample is assigned a value from a fixed set of discrete levels. The difference between the actual analog value and the nearest digital level is called quantization error or distortion. At low signal levels, this distortion can become audible as a harsh, granular noise or "birdies" – artifacts that degrade the perceived fidelity and dynamic range of the recording.

Dither solves this by adding a very low-level random noise (usually white noise or triangular noise) to the signal before the bit-depth reduction step. This noise "decorrelates" the quantization error from the signal, transforming the distortion into a constant, low-level noise floor that is far less perceptually offensive. In effect, the error becomes noise-shaped and much more like what we hear from analog tape hiss – a natural, non-distracting background. Without dither, the quantization error changes with the signal, creating a "chattering" or "modulated" distortion that is highly audible, especially in quiet passages or when fading out.

Modern DAWs apply dither automatically when exporting at a lower bit depth if the option is enabled. But the core principle remains: any time you reduce bit depth, you should dither. This includes going from 24-bit to 16-bit for CD release, or from 32-bit float to 24-bit for mastering. The noise floor introduced by dither is typically around -92 dBFS (for 16-bit) or lower, well below the threshold of hearing in most listening environments.

32-bit Float Audio Explained

32-bit float (also written as 32‑bit floating‑point) is a file format that stores audio data using a floating‑point representation rather than a fixed‑point integer representation. This is a critical distinction. In fixed‑point formats (like 16‑bit integer or 24‑bit integer), each sample is assigned a precise integer value, and the available dynamic range is fixed by the number of bits. For example, 24‑bit integer provides about 144 dB of dynamic range, while 16‑bit provides about 96 dB. These are fixed ceilings: any signal above 0 dBFS clips, and any signal below the noise floor loses resolution.

32‑bit float breaks that ceiling entirely. The format uses a 23‑bit mantissa and an 8‑bit exponent, which allows it to represent values from roughly ±1.5×10⁻⁴⁵ up to ±3.4×10³⁸ (relative to 0 dBFS). In practical audio terms, this means you can record, mix, and process signals that are many tens of decibels above 0 dBFS without clipping. When you "clip" a 32‑bit float recording, you are usually clipping the analog-to-digital converter or a plugin, not the file itself. The file simply stores the over‑level as a valid number – you can pull it down later and recover all the detail that would have been lost in a fixed‑point format.

This makes 32‑bit float ideal for:

  • Recording unpredictable sources – live concerts, field recordings, loud instruments – where you cannot set optimal gain ahead of time.
  • Internal mixing in DAWs – all modern DAWs (Protools, Logic, Cubase, Reaper, Ableton Live) use 32‑bit float or higher internally for processing. This ensures that summing hundreds of tracks and applying complex processing does not introduce clipping or rounding errors.
  • Mastering and archiving – a 32‑bit float master preserves the maximum resolution and allows future gain changes without quality loss.

The Necessity of Dithering in 32-bit Float

Now to the central question: If you are working entirely within a 32‑bit float environment – from recording to mixing to final bounce – is dithering necessary at any step? The short answer is no, as long as you never reduce the bit depth to a fixed‑point integer format. Dither is only needed when you truncate bits. 32‑bit float files are already quantized to the 23‑bit mantissa, but the floating‑point exponent means that the effective quantization step size changes with signal level. This results in a noise floor that is signal‑dependent but remains extremely low – typically on the order of −152 dBFS at 0 dBFS, rising to around −127 dBFS at −60 dBFS. These values are far below the noise floor of any practical audio system, so no audible quantization distortion is present.

However, debate arises in two common scenarios:

1. Exporting from 32-bit Float to 24-bit or 16-bit

This is a bit‑depth reduction, and dither should always be applied. Even though the source is 32‑bit float, the target is fixed‑point. Without dither, you will introduce quantization error at the new bit depth. Audio engineers sometimes mistakenly believe that because the source is high‑resolution, dither is optional – but the truncation process is deterministic and will distort low‑level detail.

2. Dithering before any processing

A more subtle point: some engineers advocate dithering before processing such as EQ, compression, or reverb when mixing in 32‑bit float. The reasoning is that internal DAW processing uses 64‑bit double precision, and dithering after each stage can prevent the accumulation of rounding errors. In practice, modern DAWs already handle this with internal noise‑shaping and are designed so that dither is unnecessary at every processing step. The 23‑bit mantissa of 32‑bit float provides approximately 24 bits of effective resolution at moderate levels, and double‑precision arithmetic (64‑bit) further reduces rounding errors to imperceptible levels. So while theoretically beneficial, the practical improvement from dithering every processing stage is negligible for all but the most extreme signal chains (e.g., hundreds of successive operations with very low‑level signals).

Scenarios Where Dithering Really Matters

Despite the theoretical advantages of 32‑bit float, real‑world workflows still involve dithering in several key situations:

  • Final delivery to streaming platforms or CD – almost all streaming services and CD players require 16‑bit or 24‑bit WAV/FLAC files. Dither is essential when exporting from your 32‑bit float master.
  • Mastering for vinyl or broadcast – these formats use limited dynamic range and require careful bit‑depth reduction. Dither helps maintain the intended sound.
  • When incorporating audio from sources with lower bit depths – if you bring in a 16‑bit file into your 32‑bit float session, that file already has quantization noise. Processing it further may expose artifacts. Dithering the 16‑bit file before gain changes can help.
  • When using plugins that output in fixed‑point – some older or hardware‑based plugins operate in fixed‑point internally. If your DAW then sends that to a 32‑bit float track, the quantization error from the plugin is already baked in. Dithering at that interface can help, but this is rare in modern plugin designs.

In most music production, the only place you need to dither is at the final export stage. Video post‑production, game audio, and sound design often involve multiple format conversions, so dither may be applied more frequently – but always only when reducing bit depth.

Best Practices for Dithering in Modern Workflows

To get the most from your high‑resolution 32‑bit float sessions, follow these guidelines:

  • Always enable dither when exporting to a fixed‑bit depth. Most DAWs have a dither option in the export/bounce dialog. Select a noise‑shaping type if available (e.g., Pow‑r, UV22HR).
  • Use the simplest dither noise for mastering. Triangular PDF (probability density function) dither is the standard – it is spectrally flat and introduces no coloration. Noise‑shaped dither moves the noise to frequencies where the ear is less sensitive, but can cause pre‑echo in some material. Trust your ears.
  • Do not dither multiple times. Dither adds noise – applying it repeatedly will accumulate. Once you reduce bit depth to 16‑bit, avoid further bit‑depth changes unless absolutely necessary.
  • Consider dithering before any significant DSP if working at 16‑bit or 24‑bit. For 32‑bit float sessions, this is unnecessary as described above.
  • Test with extreme material – a gentle fade‑out on a reverb tail or a quiet passage with low‑level detail. Listen for any “chatter” or distortion. Compare with and without dither. You will hear the difference at 16‑bit; at 24‑bit, dither is still beneficial but less obvious.

Conclusion

Dithering remains a fundamental tool in the audio engineer’s kit – but its necessity is tightly bound to bit‑depth reduction. In a pure 32‑bit float workflow, from capture to final bounce, you can safely work without dithering at any internal stage. The format’s floating‑point nature and the 23‑bit mantissa guarantee no audible quantization distortion. However, the moment you export to a fixed‑point format (24‑bit or 16‑bit), dithering is not optional; it is essential for maintaining low‑level detail and avoiding artifacts.

The debate around dithering in 32‑bit float arises because many engineers carry over habits from the 16‑bit and 24‑bit era. Understanding the technical differences empowers you to make informed decisions. When in doubt, dither – modern implementations are safe, transparent, and virtually cost‑free in terms of noise floor. But if you are staying in 32‑bit float end‑to‑end, you can confidently skip it.

For further reading, see Sound On Sound’s explainer on dithering and iZotope’s guide to dithering. For deep technical details, Bob Katz’s book “Mastering Audio” is the definitive reference (Digido – Mastering Audio). A comprehensive AES paper on 32-bit float recording can be found at AES E-Library: 32-bit Float Recording.