Dithering explained: what dither does, when to apply it, and what noise shaping adds

Dither is a small amount of noise you add on purpose when you shrink a file to fewer bits, for example from your DAW's 32-bit float mix down to a 16-bit WAV. It turns the rounding errors from that conversion into a quiet, steady hiss you will rarely notice, instead of distortion that follows the music and shows up in fades and reverb tails.

The rule that covers most of what you will do: dither once, at the very end, and only when the bit depth actually goes down. The rest of this guide explains why, what the options in your export dialog mean, and where we think artists overthink it.

What problem does dither solve?

Digital audio stores each sample as a number with a fixed number of bits. When you go down in bit depth, every sample has to be rounded to the nearest value the smaller format can hold. That rounding is called quantisation, and the error it leaves behind is not random. It is tied to the signal, so on quiet material it sounds like grit, buzz or a "crunchy" decay rather than noise.

iZotope's guide to dithering defines dither as "the intentional application of low level noise to an audio file", and notes that the distortion is most noticeable at 16 bits in reverb tails, fade-outs and quiet passages. Adding noise before the rounding breaks the link between the error and the music. You end up with a constant, smooth noise floor, which the ear tolerates far better than distortion that pumps along with the signal.

In the same guide, iZotope's own test kept a low-level sine wave audible down to nearly -120 dBFS in a 16-bit file using noise-shaped dither, which it puts at about 24 dB better than plain truncation. That is one vendor's measurement, not a universal figure, but it shows the scale of what is at stake on very quiet material.

When should you apply dither?

Apply it when the bit depth goes down, and at no other time.

• 32-bit float mix to 24-bit or 16-bit file: dither. iZotope points out that because most DAWs work internally at 32-bit float or higher, even a 24-bit WAV for mastering counts as a reduction.
• Exporting at 32-bit float: no dither needed. iZotope says that resolution produces no audible quantisation distortion.
• Bouncing stems or a mix for a mastering engineer or service: if you can deliver 32-bit float, do that and leave dither off. If you must deliver 24-bit, dither. Our stem prep guide gives the same advice.
• Final master for streaming: check what the platform asks for first (below).

Dither must be the last thing that happens

Dither only works if nothing touches the audio after it. A later gain change, limiter, fade or sample-rate conversion recalculates every sample and undoes the point of it. iZotope puts it plainly: dither "must be the absolute last edit performed on an audio file". In practice, that means the dither setting lives in your export dialog or on the last plug-in in the master chain, after the limiter.

Do it once

Apple's Logic Pro help for its bounce options says: "Avoid dithering the same audio signal multiple times." Each pass adds its own noise. One easy way to do it by accident is a limiter with its own dither switch turned on, plus dither ticked again in the export dialog. Pick one.

What does Spotify actually want?

This is where a lot of old advice goes wrong. The traditional end of a master was a dithered 16-bit, 44.1 kHz file, because 16 bits is what a CD holds. Spotify's audio file formats page asks for a sample rate of "44.1 kHz or higher", a bit depth of "24-bit if this is the native master", and "16-bit only if no higher bit depth master exists". It also says: "Don't downsample, reduce bit depth, or apply any processing that changes quality before delivery."

So if your master exists at 24-bit, send the 24-bit file and let the platform do its conversions. Your distributor may have its own spec, so check it, but don't make a 16-bit file just because that is what CDs used to need. Loudness is a separate question; our guide to streaming loudness targets covers that.

What kinds of dither are there?

You will mostly see three flavours, named after the shape of their probability distribution:

• RPDF (rectangular): noise with a uniform distribution, as Wikipedia's dither article describes it. The simplest kind; Wikipedia notes that triangular dither, below, is the type that minimises noise modulation.
• TPDF (triangular): made by adding two independent rectangular noise sources. Wikipedia notes that triangular dither minimises noise modulation, which is why it is the usual flat default.
• Shaped or coloured dither: dither whose noise has been filtered so less of it lands where hearing is most sensitive. Wikipedia says that for final distribution, "a colored dither or noise shaping is appropriate".

For a 24-bit file, iZotope says flat TPDF dither is enough, because at that depth the noise is so quiet that "at normal listening levels it's inaudible".

What does noise shaping add?

Noise shaping filters the dither and rounding noise so that its energy moves away from the frequencies we hear best and towards ones we hear poorly, mostly the very high end. The total noise does not disappear; it gets moved. iZotope compares it to an EQ applied to the dither noise, and says it matters most at 8 and 16 bits.

Logic Pro is a clear example because Apple documents what each option does in its bounce dithering help:

That help page is for Logic Pro X 10.1, so check the dialog in your version.

Which setting should you pick?

• 24-bit export: flat TPDF, or the gentlest option your DAW offers. Noise shaping buys you little you can hear.
• 16-bit export: noise-shaped dither is worth trying, especially on quiet, sparse or acoustic music with long fades.
• Heavily limited, loud masters: the music sits far above the dither, so the choice matters less. Pick one and move on.
• Material that will be processed again (stems, files for a mastering engineer): stay at 32-bit float if you can. If not, use flat dither, because shaped noise piles energy into the top end, and a later EQ boost or another conversion can bring it forward.

What we think

Dither is one of the few corners of audio where the correct answer is boring and nearly settled, and that is good news for artists. You do not need to buy a dither plug-in, A/B six noise-shaping curves, or read forum wars about which one sounds "warmer". You need to know one rule (once, last, only on the way down) and where the switch is in your DAW.

The bigger shift is that the old reason to obsess over dither, squeezing a master into 16 bits for CD, matters less for many artists now. When a platform like Spotify asks for your native 24-bit master and tells you not to reduce bit depth, the most common dithering decision becomes "send the higher-resolution file". The energy that used to go into choosing dither is better spent on the things a listener actually hears: the arrangement, the mix balance, and a master that isn't crushed to win a loudness contest normalisation will undo anyway.

Where we would still be careful is handoffs. Stems sent to a collaborator, a mastering engineer or an AI mastering service are where double-dithering and truncation sneak in, because nobody owns the last step. Agree who dithers, write it in the session notes, and deliver 32-bit float when you can. That habit costs nothing and protects the quiet details that make a record feel finished.

FAQ

Can I hear dither? At 24 bits, usually not at normal listening levels, according to iZotope. At 16 bits on very quiet passages, you may hear a faint hiss, which is the point: hiss is the trade for removing distortion.

Should I dither a 32-bit float export? No. Leave dither off when the bit depth is not going down.

My limiter has a dither switch and my DAW does too. Which one? Either, but not both. Use whichever is truly the last stage before the file is written.

Do I need to dither when I upload to Spotify? If you have a 24-bit master, Spotify asks for that file and says not to reduce bit depth, so the dithering question for a 16-bit version does not arise.

Updated 24 Oct 2026.