Bit Depth and Why It Matters
Bit depth determines the number of possible amplitude values each audio sample can have. At 16-bit, there are 65,536 possible values. At 24-bit, there are over 16.7 million. At 32-bit float, the range is effectively unlimited for practical purposes.
More bit depth means more dynamic range. 16-bit audio has a theoretical dynamic range of 96 dB. 24-bit offers 144 dB. This means 24-bit audio can represent quieter signals with greater precision, which matters during recording and processing where you want maximum headroom and minimum noise floor.
For final delivery, however, the CD standard and most streaming platforms accept 16-bit audio. This means your 24-bit or 32-bit float master needs to be reduced to 16 bits at some point. The question is how you handle those extra bits when they go away.
Quantization Noise: The Problem Dithering Solves
When you reduce bit depth by simply cutting off (truncating) the lower bits, the signal gets forced onto a coarser grid of possible values. Quiet signals that previously had smooth, gradual level changes now get snapped to the nearest available step. This rounding error creates quantization distortion.
Quantization distortion is not random noise. It is correlated with the audio signal, meaning it changes in direct relationship to the music. This makes it sound harsh, metallic, and distinctly unnatural. It is most audible during quiet passages and fade-outs, where the signal level is low enough that the rounding errors become a significant proportion of the total signal.
The irony: Truncation distortion sounds worse than random noise of the same level because your brain can detect signal-correlated artifacts far more easily than uncorrelated noise. Dithering exploits this by trading correlated distortion for uncorrelated noise, which is perceptually less objectionable even when measured at a similar level.
At normal listening levels on a well-mastered track, truncation artifacts are subtle. But in quiet passages, during long fade-outs, and on high-quality playback systems, they are audible. Professional mastering does not leave audible artifacts in the final product.
What Is Dithering?
Dithering adds a very small amount of carefully designed random noise to the audio signal before the bit depth reduction occurs. This noise randomizes the rounding errors, converting the correlated quantization distortion into uncorrelated random noise.
The result is that instead of hearing metallic, signal-dependent artifacts in quiet passages, you hear an extremely faint, constant hiss. This hiss is at a far lower level than the music and is virtually inaudible during playback. It is the lesser of two evils, and in practice, it is no evil at all.
The concept seems paradoxical: adding noise to improve quality. But the math is clear. Correlated distortion at -96 dB sounds dramatically worse than uncorrelated noise at -93 dB. Dithering trades a small, measurable degradation in signal-to-noise ratio for a large, audible improvement in perceived quality.
TPDF Dither: The Industry Standard
TPDF stands for Triangular Probability Density Function. It is the standard dithering algorithm used in professional mastering and is the type recommended by the AES (Audio Engineering Society).
TPDF dither uses noise with a triangular amplitude distribution, generated by summing two independent rectangular noise sources. The key property of TPDF dither is that it completely eliminates quantization distortion without adding any noise modulation. The noise floor remains perfectly constant regardless of the signal level.
Other dither types exist but are less ideal for mastering:
- Rectangular (RPDF) dither: Uses flat-distribution noise. Simpler than TPDF but leaves some signal-correlated noise modulation, meaning the noise floor subtly shifts with the music. Not recommended for final mastering.
- Gaussian dither: Uses noise with a bell-curve distribution. Slightly higher noise level than TPDF with no practical benefit. Rarely used in audio.
- Shaped dither (POW-r, UV22, etc.): TPDF dither with noise shaping applied. These are the most common choices in professional mastering. See the next section.
Noise Shaping: Hiding the Noise Where You Cannot Hear It
Noise shaping is an additional processing step that redistributes the dither noise across the frequency spectrum. Instead of spreading the noise evenly from 20 Hz to 22 kHz, noise shaping pushes most of the noise energy into the high-frequency range above 15 kHz, where human hearing is least sensitive.
The total noise energy remains the same (or slightly increases), but the perceived noise drops significantly because your ears are far less sensitive to frequencies above 15 kHz. The result is a quieter-sounding master with the same mathematical noise floor.
Common noise shaping algorithms include:
- POW-r Type 1: Gentle noise shaping. Good for all material. The safest choice when you are unsure.
- POW-r Type 2: More aggressive shaping. Lower perceived noise but pushes more energy to the extreme high end. Works well for music with lots of high-frequency content (cymbals, strings, acoustic instruments).
- POW-r Type 3: Maximum shaping. Lowest perceived noise but can occasionally produce artifacts on very quiet material. Best for dense, loud genres where the noise floor is irrelevant during playback.
- UV22 (Apogee): A proprietary algorithm that injects a single ultrasonic tone rather than broadband noise. Effectively inaudible but adds energy at specific high frequencies.
For most mastering work, TPDF with moderate noise shaping (POW-r Type 1 or Type 2) is the correct choice. It eliminates quantization distortion while keeping the noise floor well below audibility on any playback system.
When to Dither (and When Not To)
The rule is straightforward: apply dithering whenever you reduce bit depth. Here are the common scenarios:
- Exporting 24-bit or 32-bit float to 16-bit WAV: Always dither. This is the most critical case because you are losing 8 or 16 bits of resolution.
- Exporting 32-bit float to 24-bit WAV: Technically you should dither, but the noise floor at 24-bit (-144 dB) is so far below audibility that truncation artifacts are inaudible. Many engineers skip dithering here with no practical consequence.
- Exporting to MP3, AAC, or Ogg Vorbis: The lossy codec applies its own noise management during encoding. Adding dither before lossy encoding is unnecessary and slightly counterproductive because it adds noise that the codec then has to encode.
- Bouncing at the same bit depth: No dithering needed. If your source is 24-bit and your output is 24-bit, there is no bit-depth reduction.
- Applying dither multiple times: Never dither more than once. Dither should be applied exactly once, at the very last stage of the signal chain, right before the final export. If you dither during mixing and again during mastering, you have added noise twice for no benefit.
Placement rule: Dithering must be the absolute last process in your mastering chain. Any processing after dithering (EQ, compression, limiting, gain change) will undo the dither's statistical properties and reintroduce quantization distortion. The chain order is: all processing, then true peak limiting, then dithering, then export.
LuvLang handles dithering automatically during export. When you download a 16-bit WAV master, noise-shaped TPDF dither is applied as the final processing step, ensuring your deliverable is free of quantization artifacts without any manual configuration. For 24-bit and lossy format exports, dithering is managed appropriately for each format.