Tracking/setting levels: where do your tracks peak?
- Thread starter J.DavisNJ
- Start date
I think ideally setting levels for each instrument with plenty of headroom makes the most sense. Once you start adding up tracks and bussing and all that jazz your going to be turning all your faders down just for headroom to mix. If you track instruments evenly you've already got a decent "mix" before you even start mixing.
It's not my fault. I could see Fart666 or Felch666.
I'm with you on that one. If you have to pull down the faders on everything, you are essentially fucking with the bits to get the levels to drop. So get em as close to proper on the way in, and you won't have to worry about all that.
John_C
formerly Skeksis268
I'd just like to clarify that going all the way to 0 does result in better sampling.
for example, at 8 bit, using two's compliment, a sample at 0db would be represented by 01111111 or 10000000, depending on whether it is fully positive or negative
this continues in equal voltage steps until complete digital silence: 00000000
if you track at -12 then a certain percentage of the available values of each sample will never be used. You are limiting the number of sample values between your peak volume and complete silence, meaning an effective reduction in bit depth.
However, the fact that converters are calibrated at lower than 0db probably negates any bit depth advantage gained by tracking that hot. Also, there's nothing as as irritating as having a vocalist do a perfect take and finding that it's clipped because they leaned into the mic a bit more than during setup.
However, it may be that the converters are in fact calibrated at lower RMS values as opposed to peak values it would mean that it would definitely be better to track hot as you'd want your RMS value to be averaging around the calibration point for best results.
for example, at 8 bit, using two's compliment, a sample at 0db would be represented by 01111111 or 10000000, depending on whether it is fully positive or negative
this continues in equal voltage steps until complete digital silence: 00000000
if you track at -12 then a certain percentage of the available values of each sample will never be used. You are limiting the number of sample values between your peak volume and complete silence, meaning an effective reduction in bit depth.
However, the fact that converters are calibrated at lower than 0db probably negates any bit depth advantage gained by tracking that hot. Also, there's nothing as as irritating as having a vocalist do a perfect take and finding that it's clipped because they leaned into the mic a bit more than during setup.
However, it may be that the converters are in fact calibrated at lower RMS values as opposed to peak values it would mean that it would definitely be better to track hot as you'd want your RMS value to be averaging around the calibration point for best results.
But when you have a max range of 144 dB at 24-bit resolution, you can't seriously believe only making it even to like 120 dB (AKA -24 I assume, unless I'm missing something) will make any sort of noticeable difference, especially with how low a noise floor most gear has these days; as I mentioned, I'd be more concerned with pushing the analog stages of prosumer pres and converters (but also, to be clear, I doubt that would make more than like a 3% difference in the sound quality of the final product, but hey, this whole thread is about nitpicking anyway, so why stop now
)
John_C
formerly Skeksis268
But when you have a max range of 144 dB at 24-bit resolution, you can't seriously believe only making it even to like 120 dB (AKA -24 I assume, unless I'm missing something) will make any sort of noticeable difference, especially with how low a noise floor most gear has these days; as I mentioned, I'd be more concerned with pushing the analog stages of prosumer pres and converters (but also, to be clear, I doubt that would make more than like a 3% difference in the sound quality of the final product, but hey, this whole thread is about nitpicking anyway, so why stop now
It's not the dynamic range i'm really talking about being a problem, it's the resolution within that range
I don't follow; the amount of bits that are 1's (or the size of the binary number) simply determines the amplitude of the signal, so the resolution won't be affected by the level of the signal; it's not like bits aren't being used at lower levels, they're just smaller values (more 0's)
SethTheGreat
Member
There's a really solid discussion on this topic here:
http://www.gearslutz.com/board/so-m...s-don-t-sound-good-analog-mixes-restored.html
http://www.gearslutz.com/board/so-m...s-don-t-sound-good-analog-mixes-restored.html
Ermz
¯\(°_o)/¯
That's understandable, but shouldn't really matter much since the end product will likely be downsampled to 16-bit. You can effectively track at -50dB and still be getting the same bit depth as your end product. Not that it would be advisable... nor sane, but all the same it illustrates to not get too caught up in this.
If you have converters that are properly calibrated (ie. not RME... sigh) you'll have 0dBVU corresponding to roughly -18dBFS. 18dB of headroom, corresponding to the headroom of good, high-end mixing consoles. So if all your outboard were calibrated for this, you would track the bulk of your steady state signals around -18, with your transient peaks shooting up to -12 or whatever.
I tend to work a little bit edgier, with the bulk of signals around -12dB and the peaks hitting around -6. I find I brickwall most of the time while I'm mixing, so it just makes more sense to me.
John_C
formerly Skeksis268
Thanks for the explanation and the info on calibration. The downsampling thing is important, i forgot about that. The only thing i can think of now is that plugins will work better when given a better resolution to work with, but it's all getting a bit hypothetical now.
this is an extreme example as the percentage of the information lost is huge. However, the principle still holds at any bit depth. As ermz said though, good converters are calibrated for steady signals peaking at about -18 and the errors there are likely dominant over the rounding error when you're working at 16+ bits
edit: my first piece of forum mspaint
I'm still not following though; again, the value of the binary number for each sample simply indicates its amplitude, or position within the overall dynamic range of the current bit depth, so it's pretty much like graph coordinates for a point - thus, all the points are assembled, and the wave is created by the D/A playing connect-the-dots between 'em (and the sample rate of course determines how frequently samples are taken), and as far as I understand having louder signals will just result in a bigger range of binary numbers between the peaks and valleys, though no difference in resolution?
And I'm pushing this just because I'm really curious about the subject, currently taking Computer Architecture atm
And I'm pushing this just because I'm really curious about the subject, currently taking Computer Architecture atm
Oh wait, I think I'm getting it - because the range between largest and smallest numbers is reduced with a lower signal, there isn't as much detail between those because your precision is limited by the size of the binary word (similar to using a float vs. double in C++); however, when you're working with 24-bit words (and thus have 2^24 possible values), and the difference is whether you're peaking at -20 (or even -40) or 0, jesus fucking christ, the difference has gotta be like a fraction of a percent 
(I'm sure it would be possible to figure it out, but I'll pass )
(I'm sure it would be possible to figure it out, but I'll pass )
Haha, nah, when it comes to audio engineering all this digital audio stuff is firmly rooted in the realm of the general rather than the applied, so IMO unless you have aspirations to build converters or something it's not crucial info
John_C
formerly Skeksis268
Oh wait, I think I'm getting it - because the range between largest and smallest numbers is reduced with a lower signal, there isn't as much detail between those because your precision is limited by the size of the binary word (similar to using a float vs. double in C++); however, when you're working with 24-bit words (and thus have 2^24 possible values), and the difference is whether you're peaking at -20 (or even -40) or 0, jesus fucking christ, the difference has gotta be like a fraction of a percent
i think you've got the right idea
I did some maths and I came up with examples:
if you're recording at 16 bit with peaks at roughly -12dBFS you are effectively recording at 14 bit as far as the effective dynamic range and amplitude resolution is concerned. That means that in fact you've only got 1/4 of the resolution (it looks bad, but actually it'd be quite hard to hear)
@ 24 bit, if you track with peaks at -50dBFS, you are effectively working in 16 bit.
Again, what might appear big mathematically (2^24 compared to 2^16) is in fact audibly pretty small and no one in their right mind would track at -50!!!!!!
how embarrassing that i had to do all that work and ermz already had the same result
To make my life easier, I have my meters in Cubendo setup so that I'm in the green all the way up until around -12, yellow goes to -6 and above that is RED. No fading like the default meters do...
pretty much, meters go from straight GREEN, to straight yellow, to straight RED. As long as I'm in the upper green, occasionally hitting the yellow, I'm golden.
I recommend this to anyone using Cubendo...
-18db is the ideal range to record in the digital domain, IMO. All analog consoles are gear are basically calibrated for this and hence, most analog modeling plugins are as well....
... or something like that. technicalities aren't fun enough for me. =D
pretty much, meters go from straight GREEN, to straight yellow, to straight RED. As long as I'm in the upper green, occasionally hitting the yellow, I'm golden.
I recommend this to anyone using Cubendo...
-18db is the ideal range to record in the digital domain, IMO. All analog consoles are gear are basically calibrated for this and hence, most analog modeling plugins are as well....
... or something like that. technicalities aren't fun enough for me. =D
