Galen writes:
“Dither and Sample rate conversion question-
Reading Bob Katz’s book, mastering audio.
I want to clarify- That I have heard from 2 independent sources, the Bob Katz book, and Dan Worrall, that one should use 24 bit dither when bouncing to 24 bits, because it is a wordlength reduction, coming from the daw’s 64 bit floating point. If I am incorrect I would like to know but I believe those of you saying dither only when converting to 16 bit are incorrect. If I am misunderstanding Dan Worrall and Bob Katz please correct me.
I have a project which I was mixing, the file is 24/96.
I bounced the project to one stereo file, also 24/96, and used goodhertz’s good dither as the last part of the chain on the output, to dither at 24 bits.
Now I am mastering this song, and I am trying to clarify in what order I should ideally do SRC to 48khz and 24 bit dither, to get 24/48 master.
I am reading in the book that Dither should be the last stage, even after SRC.
In Izotope RX8 I can do SRC and also host audio units, so I was thinking I could do SRC in RX, then follow with a 24 bit dither, then export directly from RX. But I have mostly exported from Logic.
Anyway, if there’s something else I should be doing ideally, let me know.”
Dear Galen
Hi, Galen. My book explains that wordlength grows with any multiplication operation. SRC uses a ton of multiplications, and the SRCs on the market are all floating point and their output wordlength is mostly 32 bit float with some 64 bit. No matter what source wordlength, e. g. 16 or 24 bit, you’ll get 32 float out of the SRC, or even a simple gain change in a floating point DAW.
So, if possible, leave it at 32 bit float till you are done with ALL the operations in line, until the final wordlength reduction.
How audible is a truncation to 24 bit? Let’s just say “audiophile-level audible”, which is quite subtle. Typically Some slight loss in depth or soundstage width. People say, if that’s the case why bother. No one’s gonna notice.
But keep in mind that distortion is cumulative. A little loss here and there can add up, particularly if the end product is lossy coded. And being technically correct doesn’t take much brain power once you learn the simple rules.
Lastly, I will say that going to properly dithered 24 bit as an intermediate step is not sonically costly at all, as the dither noise at -141 dBFS is inconsequential. So if you have to deliver, say, a 2444 master to some service, and also a 1644, I wouldn’t worry about using the properly dithered 2444 as an intermediate product, making the 1644 from the 2444, there’s no hill to die on there.
If you ever do a null test between the 24 bit and a 16 bit result, 9 times out of 10 the null residual will only be dither noise, no musical information or distortion. What’s that all about? Just noise? But that 16 bit dither noise is usually enough to subtly change the depth or tonality by virtue of masking. So choose your dither shape wisely! It’s also the main reason why I avoid making mp3 or AAC refs from a 16 bit source when possible, and go back to the 24 or even the 32 float to create the mp3. Because the tonal shift, the depth losses are even more audible at the lossy stage. Always consider that there are going to be more stages and manipulations to the product AFTER it leaves your hands!