16 vs 24 bit: which master to deliver

"What should I deliver the master in?" sounds like a simple question. Around bit depth there is a lot of belief and very little arithmetic: some treat 24-bit as a badge of quality, others as marketing. Here is what actually sits behind it.

What bit depth really is

Bit depth is not "sound quality" and not "density". It is the number of steps used to describe the amplitude of each sample. From that number follows exactly one property: how quiet a signal can be and still stand apart from rounding noise.

The arithmetic is simple, roughly 6 dB per bit:

The key point: those 96 or 144 dB describe where the noise floor sits, not how "detailed" the audio is. Quiet passages in 16-bit do not "lose detail" — they sit 96 dB above the noise, which is ample for a finished master.

Why you will not hear the difference

The noise floor of a dithered 16-bit file is around −93 dBFS. To hear it, the silence in your room and the self-noise of your amp and headphones would have to be quieter still. Real rooms rarely sit below 30 dB SPL, and to bring −93 dBFS noise into audibility you would have to turn up to a level where the music itself is deafening.

So the honest answer: on a finished master, at normal listening levels, 16-bit and 24-bit are indistinguishable. People who claim otherwise have usually compared files at different loudness or from different masters.

So why 24-bit

Because a master is not the end product for a listener — it is the source for everything downstream.

• Platforms transcode it into their own lossy formats. The encoder works with what it is given; hand it 16-bit and it starts from material that already carries one generation of quantisation noise.
• Albums get re-sequenced, tracks get reissued, radio edits and remasters happen. Every depth reduction adds a generation.
• If the recipient needs to match levels or assemble a compilation, 24-bit has headroom to give. 16-bit does not.

It is the same logic that makes photographers deliver a processed high-quality file rather than the camera JPEG. The difference shows up not on first view, but on the second round of processing.

Dithering: required going down, harmful otherwise

This is where confusion usually lives. Dithering adds very quiet noise when reducing bit depth. The reason: rounding error is correlated with the signal, so it is heard as distortion rather than noise. Dither breaks that correlation — the error becomes flat noise, which hearing forgives far more readily.

Two rules follow:

1. Going down — dither. From 32-bit float to 16-bit, always. The noise type decides where it lands spectrally: TPDF spreads it flat, noise-shaped variants (ns_e, ns_itu) push it where hearing is less sensitive.
2. Not going down — do not dither. Exporting 24-bit from float leaves rounding error around −144 dB anyway. Adding noise there degrades the signal for the sake of ritual.

Magic Master keeps these separate: the 16-bit export carries dithering with a choice of type, the 24-bit export carries none, because none is needed.

What to pick in practice

• Distribution, streaming, delivering to a client — 24-bit. WAV if WAV is requested; FLAC if size matters (about half, at identical quality).
• CD — 16-bit with dithering. The format requires it.
• Sending a friend a listen, messenger uploads — 16-bit, or plain MP3 320. Here 24-bit only inflates the file by half with no audible upside.

"By half" is not a figure of speech: a four-minute stereo track at 44.1 kHz is about 42 MB in 16-bit and about 63 MB in 24-bit. For a single master that is nothing; for an album over mobile data it is not.

How to do it in Magic Master

The export block offers WAV 16-bit — free and selected by default, with a dither type selector. WAV 24-bit and FLAC 24-bit come with token mastering: both are distribution formats with identical results, FLAC simply weighs half as much. The mastering itself is identical — processing runs in 32-bit float and depth is chosen only at export, so "24-bit" makes nothing louder or denser. It simply preserves the result without an extra quantisation generation.

If in doubt, take 24-bit. It spoils nothing and the headroom may be needed. The only cost is file size.

In short

• Bit depth sets where the noise floor sits, not how detailed the audio is.
• On a finished master, 16 and 24 bit are indistinguishable: a −93 dBFS floor is below any real room.
• 24-bit matters as a source: transcoding follows, and every depth reduction adds a generation of noise.
• Dither only when reducing depth. In 24-bit it is harmful.
• WAV and FLAC at 24-bit are identical in quality; FLAC is half the size, WAV is accepted everywhere.