Just wrote this out and thought I should post it on here… probably posted the same kind of thing before but never mind.
***
Q: “Should I be bouncing [at a] higher [sample rate/bit depth]?”
Re bit depth / sample rate… Quick explanation.
Terminology thing first: bit rate is different to bit depth. Bit rate = bit depth x sample rate. So CD audio is 16 x 44100 = 705600, x2 cos it’s stereo = 1411200 = 1411kbps. Which is over 4x as much data per second as a 320kbps mp3.
[EDIT: This last bit was edited – see comments. This doesn’t mean the 320kbps mp3 has one quarter of the signal. I forgot about the data compression side of things when positing originally, apologies]
Anyway, bit depth, always export at 24-bit basically. Lower noise floor, ie better precision/quality at low levels. Rule of thumb is 6dB dynamic range per bit. So 16 bit audio has its noise floor at 16 x 6 = -96dBFS. 24 bit = 24 x 6 = -144dBFS. You might have seen people say about ‘the stair steps getting smaller’ which is incorrect; they just go further down.
So 24-bit means any subsequent processing is done from a better starting point, and hence is a good thing.
Sample rate: if you are going to work at higher sample rates then you must do it from the start. There is zero point in mixing at x rate and then exporting to a higher one. So everything you record from your outboard/analogue, do it at the higher rate, mix at the higher rate, export at the higher rate, I’ll master at the higher rate and then bring it down at the end.
Underlying point here is that at some point it will have to come down again, and sample rate conversion is a pretty nasty thing. Mine is very good here, but I’d still rather avoid it. It’s fine to send out to analogue at one rate and capture at another but it obliges a certain converter combination – [unless you have] a million converter options.
Blah blah blah. TL;DR = either do it from the start and stay at higher sample rate the entire time, or don’t worry about it.
IMHO the benefits of working at 48k are at least balanced by the downsides of downsampling to 44.1k, so, whatever. If you do choose to do it, 88.2k is plenty, 96k might even be worse (search for Dan Lavry’s famous white paper if you are really really bored).
But all this is a comparative side-issue if you have 200dB too much sub bass, you know?
Boring enough for you? Good 🙂
Bob Macc's Blog 
#1 by forestatq on January 22, 2018 - 2:53 pm
Hey Bob, is there ever any reason to export 32 bit audio?
#2 by bobmaccsblog on January 22, 2018 - 5:59 pm
32-bit floating point is practically identical to 24-bit fixed point. The extra 8-bits allow it to be a ‘sliding scale’, providing something like 1700dB (yep) headroom above 0dBFS. This is why you can clip your DAW channels/busses internally; they run at 32-bit FP. The real world is fixed point though, so you have to come back under 0dBFS anyway to avoid clipping your converters.
Soooo, to answer the question, if your mix clips and you export it to 32-bit FP then you can get all the clipped information back by turning it down. But if you need to do that, you were monitoring a mix that was clipping the whole time. So the turned-down mix will sound different to what you were listening to. So just don’t clip the feckin mix.
In summary: not really in the overwhelming majority of cases.
🙂
(n.b. skipped a load of stuff here, and was trying to play Castle Crashers with some kids)
#3 by Forest Bond on January 22, 2018 - 9:26 pm
[sorry if duplicate comment, having a hard time with comment system]
OK makes sense — but what about dithering? I.e. since the DAW is using 32 bit float internally, when exporting to 24 bit we would want to do some dithering, right? But then the mastering engineer would also apply some dithering when moving down to 16 bit. Or do we just assume that dithering once at mastering is sufficient?
#4 by bobmaccsblog on January 22, 2018 - 10:49 pm
A few things overlapping here.
“since the DAW is using 32 bit float internally, when exporting to 24 bit we would want to do some dithering, right?”
Nope – remember that practically speaking, 32-bit FP is the same as 24-bit fixed (plus 8 bits for the exponent, the sliding scale). No need to dither because you aren’t reducing the bit depth.
Take your time and look at fig 1.1 and 1.2 here. I’ve read this paper loads of times and get about 0.1% closer to truly understanding it all every time…
Click to access floatingdither.pdf
“the mastering engineer would also apply some dithering when moving down to 16 bit. Or do we just assume that dithering once at mastering is sufficient?”
Dithering is necessary when you reduce bit depth (word length in the terminology of the paper above). In reality you probably have lots of dither stages in a mix – in your plugins that process at higher internal bit depths etc. But yeah, as far as yer normal producer-type goes, just export at 24-bit and don’t dither.
Does this help? Long day…
#5 by Forest Bond on January 23, 2018 - 3:25 pm
It does help, I think, but I’m gonna need a minute or two with that paper. Thanks a bunch for the insight.
#6 by Forest Bond on January 23, 2018 - 7:00 pm
Yeah I dunno, having a hard time proving to myself that (a) there is no quantization error when exporting to 24 bit and therefore (b) there is no risk of quantization distortion and therefore (c) there is no need for dithering. I’ve read somewhere that the quantization distortion would be so small as to be inaudible, which is easier for me to accept… but I’ll probably still ship 32-bit pre-masters. 😀
Disclosure: I definitely did not fully understand that paper and I don’t feel like I really get 32-bit float audio representations.
#7 by duffah on January 22, 2018 - 6:16 pm
Good explanation of bit depth Bob. Almost like the depth refers to the depth (or how low) of the noise floor.
#8 by bobmaccsblog on January 22, 2018 - 7:30 pm
Shocking innit.
#9 by antou on January 23, 2018 - 5:38 pm
> Which is over 4x as much data per second as a 320kbps mp3. So the best mp3 throws away more than three quarters of the info…
Wow, this is so wrong. This is like saying FLAC throws away a portion of the information because FLAC files are smaller than WAV.
MP3 compresses data by, for example, taking advantage of signal redudancy. It is a lossy codec, so it does throw away some of the original signal, but certainly not three quarters of it.
You probably know this already, but other people might just remember “4x smaller bitrate/filesize = 4x less signal”
#10 by bobmaccsblog on January 23, 2018 - 5:52 pm
Completely right, of course – I’ll edit to reflect that. In fairness I was typing at high speed while tired and it was something of a side point. But thank you 🙂
However it does raise something I’ve always wondered about, and never bothered to look into. How much signal (approximately) actually does get thrown away before you consider the data compression side of things? I’m not an expert on that stuff *at all*, but to take a simple example; same track at 24-bit vs 16-bit has a much better RAR compression ratio. I only ever use RAR so dunno about anything else. This is because of ‘all the zeroes’ (to bastardise the concept), presumably. So reducing the amount of signal via perceptual encoding would reduce the amount of available data compression, right?
Do you have any information on that? I’d be interested. Again, I’m not an encoding/compression expert. Maybe it doesn’t work like that at all, ha. I am very happy to get pointed to some reading etc. I’ll read it when I’m not drowning in work…
Cheers!
#11 by bobmaccsblog on January 23, 2018 - 7:17 pm
On a completely unrelated note, bloody hell, there’s actual conversation happening on my blog. I didn’t even know anyone actually read it :’)
#12 by Daniel on January 17, 2019 - 11:53 pm
I read it regularly! I’ve learned a lot from your posts and I really enjoy your style of writing. I know I’m not the only one, so just know that you’ve got some loyal readers out here (even if we’re not commenting).
#13 by bobmaccsblog on January 18, 2019 - 7:27 am
Awww 🙂