Found a PDF on digital audio levels/considerations for gain staging and mastering

[porchy]

Hi all,

Stumbled upon this today. It's a bit dated but still has some interesting high-level information while offering a more tangible approach on how to "view" certain things in your mixer. Additionally, the author, Holger Lagerfeldt, included some incredible links at the bottom of this concise runthough.

http://www.popmusic.dk/download/pdf/levels-in-digital-audio.pdf

Hope you guys find this as useful as I have!

-Anthony

 

[Drew]

This is one of the better reads on gain-staging I've seen:

Proper Audio Recording Levels | Rants, Articles

Great advice too. It's VERY hard to .... up a mix by recording too quiet, but it's incredibly easy to do so by tracking too hot. Tracking hotter than you need to without actually clipping, and then turning levels down to prevent your output bus from clipping works, but why give yourself all that extra work?

 

[vansinn]

Quite good read (the top posted PDF), and thanks for the link
A couple of comments:

First one is a Bit semantic
He states that recording at 24 bits yields more resolution, which is important especially at low levels This isn't wrong per se, but can be misleading.
Obtaining a noise floor at some 70-80 dB down in a project studio is quite a task, meaning those lower bits will easily be buried in the noise below the noise floor.
Granted, there's still information down there, but it's less practically useful.

Next, I partly disagree on recording at -3 dBFS, as all it takes is i.e. the sax player feeling extra horny and blowing just a tad more, or a singer maike a more pronounced 'P' type of puff to have an overload.
Especially when recording in 24 bits, there's ample of space in both directions to record with good fidelity at a more safe -6 or more dBFS.

EDIT: Interesting that the send post says the same about recording too hot

Towards the end, he says applying dithering is always a good idea when going from higher to lower bit levels.
I'd say it should always be applied.


A topic he doesn't touch is sampling frequencies.
I'm not going to start a discussion on whether 96 kHz is better than 48, but just want to mention that many plugins are programmed for a specific sample rate - often 48 kHz, and internally does up- and down scaling incl. dithering.

This means that, if/when using such, the 96 kHz samples one might have started with won't be maintained as expected.
Rather, while the end result will of course be 96, it will have been changed underways.

You might be surprised to know how many studios will record and process in 48 kHz, and simply upscale to 96 or 192 when the client wants to check the progress or add something, and then downscale to 48 again when they get it back. The client will receive the final result upscaled to 96/192 and perfectly happy

This doesn't mean I object to recording/processing at higher rates, just intended as notes..


The second link is a decently better read, by someone who really knows, thanks

 

[tedtan]

The norm that most pro studios are working at today is a level of -18 dBFS average, which puts the peaks between -12 and -6 dBFS. This approach has several advantages, and I would recommend it to anyone working in 24 bit (and if you're not working with 24 bit, start working in 24 bit).

First, and probably most importantly, -18 dBFS is equivalent to 0 dB on an analog VU meter, and is the level at which analog gear is designed to work. And you don't even have to use analog compressors, EQ's, summing boxes or mixers to benefit - a lot of plugins are designed to emulate analog gear and they are often coded so that they work best in this range, too. If pushed too hard, they sometimes clip in an unpleasant way.

Second, when you use certain effects, you end up add volume to your track (sometimes without realizing it). EQ, distortion, saturation, compression etc. often do this (any gain based effect can), which when pushed too far will cause digital clipping. By leaving a decent amount of headroom (18 dB), you have plenty of room to tweak without accidentally clipping.

 

[tedtan]

Towards the end, he says applying dithering is always a good idea when going from higher to lower bit levels.
I'd say it should always be applied.

Dithering is used to cover up the noise of quantization distortion when you go from a higher bit depth to a lower bit depth. Many people question whether it benefits anything, but I agree that it should be used.

But when going from a lower bit rate to a higher bit rate (say 16 bit to 24 bit), it is not needed because you still have all 16 bits of data - you just have 8 more bits of 0 of lower significance added, too.

 

[callankirk]

Towards the end, he says applying dithering is always a good idea when going from higher to lower bit levels.
I'd say it should always be applied.

Agreed, agreed, agreed x 100. I apply dithering regardless of bit level conversion. IMO, what can it hurt???

A topic he doesn't touch is sampling frequencies.
I'm not going to start a discussion on whether 96 kHz is better than 48, but just want to mention that many plugins are programmed for a specific sample rate - often 48 kHz, and internally does up- and down scaling incl. dithering.

This means that, if/when using such, the 96 kHz samples one might have started with won't be maintained as expected.
Rather, while the end result will of course be 96, it will have been changed underways.

You might be surprised to know how many studios will record and process in 48 kHz, and simply upscale to 96 or 192 when the client wants to check the progress or add something, and then downscale to 48 again when they get it back. The client will receive the final result upscaled to 96/192 and perfectly happy

In my experience, this is somewhat the "norm" in big studios to allow for better real-time processing and then better quality (i.e. higher bit rate) processing upon mixdown.

Good article, too! And good points to add!

 

[4Eyes]

I saw similar discussion on Andy Sneap's forum - so how to correctly set up levels for recording DI guitar for example? general idea is to record at -18dbfs (RMS) = 0db on VU meter.

till now I've always set up gain on instrument preamp to not hit 0db on VU meter in Reaper at highest peak (pinch harmonics on low strings in my case). when I put meter on DI track recorded in that way it was around -25dbfs (RMS) / -7db on VU when playing chords over all 6 strings. According to -18dbfs rule I should increase gain to have input signal slightly under 0db on VU when playing chords over all strings, but in that way I'll have some clipping when playing pinch harmonics on low strings. by clipping I mean hitting 0.0db on VU meter in Reaper, I didn't noticed it would be more.

what's correct - setting gain level to value when it reaches slightly under 0.0db on VU meter at highest peaks (pinch harmonics) or to set up gain level to value when general playing is slightly under 0.0db on VU meter and having some clips on VU meter during peaks is ok?

 

[vansinn]

Originally Posted by vansinn
Towards the end, he says applying dithering is always a good idea when going from higher to lower bit levels.
I'd say it should always be applied.

Dithering is used to cover up the noise of quantization distortion when you go from a higher bit depth to a lower bit depth. Many people question whether it benefits anything, but I agree that it should be used.

Yep, and there's a Bit more to it - though we're likely talking about the same..
The bits below what's intended to be left behind after the reduction, i.e. the ones below bit 16, all contains valid info.
Simply truncating from 24 to 16 isn't at all the same as calculating 'up' the lower bits into a resulting 16 bit word, which is part of what dithering does.
Depending on how much info was in the lower bits, this can make difference between the resulting 16 bit sample ending on i.e. 01 or 11.

Now, if we assume a 24 bit sample has information down to, say, 17 or 18 bits, with the remaining bits being zero, we may not be able to tell any difference between a simple truncation vs a dithered scaling - at least not if done once only (like on a single source).

But if this is repeatedly done on any sources carrying info further down the bit levels, it most certainly will start to make a difference.

It's my understanding that by far most (all?) DAW's automatically does dithering instead of simply truncation, but I could be wrong. Maybe it's only done (auto) right in some situations..
At least I know this used to be much discussed topic up until some years ago..

But when going from a lower bit rate to a higher bit rate (say 16 bit to 24 bit), it is not needed because you still have all 16 bits of data - you just have 8 more bits of 0 of lower significance added, too.

Yep.

 

[vansinn]

I saw similar discussion on Andy Sneap's forum - so how to correctly set up levels for recording DI guitar for example? general idea is to record at -18dbfs (RMS) = 0db on VU meter.

till now I've always set up gain on instrument preamp to not hit 0db on VU meter in Reaper at highest peak (pinch harmonics on low strings in my case). when I put meter on DI track recorded in that way it was around -25dbfs (RMS) / -7db on VU when playing chords over all 6 strings. According to -18dbfs rule I should increase gain to have input signal slightly under 0db on VU when playing chords over all strings, but in that way I'll have some clipping when playing pinch harmonics on low strings. by clipping I mean hitting 0.0db on VU meter in Reaper, I didn't noticed it would be more.

what's correct - setting gain level to value when it reaches slightly under 0.0db on VU meter at highest peaks (pinch harmonics) or to set up gain level to value when general playing is slightly under 0.0db on VU meter and having some clips on VU meter during peaks is ok?

I'd say set it to 0 dB using the strongest source, i.e. when playing in your most attacking style

 

[Alphanumeric]

Hmm I record guitars at around -14 to -12 dbfs, though this may be too much but this seems to say you can go to -12. However I record into my mastering chain and I do a lot of processing on the fly with drums first.

Guitars are a little different when recording straight from something like the axe fx/pod where no real preamp/conversion stage is concerned. It means a lot less just to a DAW, and say, some amp sims.

There's a danger here that people who just do ITB and have no preamps or converters are gonna end up recording at a level meant to be for NOL on analog gear, and to bring their levels up again when mixing they are going to upward compress too much/overcompress and limit, I've seen this before.

Good articles. Gain staging is important. Basically, don't record hotter than the level your damn stuff is designed for duuuhhh

Of course, this is standard practice. However for metal, clipping converters during tracking is quite a popular thing for getting some tasty sounds and running things to get more volume without worrying about dynamics (when was the last dynamic metal record?). Very popular with toms and snares. Anything with a big attack. Not so much for constant level like guitars. I've seen this described as an analog version of softwear soft clipping when levels are brought back down again when mixing.

 

[tedtan]

what's correct - setting gain level to value when it reaches slightly under 0.0db on VU meter at highest peaks (pinch harmonics) or to set up gain level to value when general playing is slightly under 0.0db on VU meter and having some clips on VU meter during peaks is ok?

If it's an actual VU meter or a digital simulation of an analog VU meter, you can go over 0. It will still have 18-20 dBFS worth of headroom.

If you're using a meter that measures peaks, definitely keep the highest peaks below 0 dBFS. I'd probably make sure you keep the peaks between -3 and -6 in that situation just in case you hit one of those harmonics a little harder than usual.