Digital gain increase question

From Edison cylinders to pre-amps to ProTools: talk about it here.
User avatar
Crummy Old Label Avatar
Posts: 1226
Joined: Wed Aug 25, 2004 5:55 pm
Location: Out of my fucking mind

Digital gain increase question

Postby Crummy Old Label Avatar » Mon Dec 26, 2005 6:46 pm

Say you have a digital sound file that peaks at -12.56 dB. Now, say you use your editing program to increase the gain +12.55 over the enitre file. The file now peaks at -0.01 dB.

My question is this: why do some consider this to be a very bad thing thing to do to a sound file? By making sure that the peak remains under 0 dB -- and assuming the WAV editor doesn't use screwy algorithms, I can't understand what the harm could possibly be. Yes, you will increase the noise floor, but you'd do that anyway by having to crank up the listening volume of such a quiet track to begin with.

Common sense tells me that it would be best to increase the gain in the digital domain first, rather than have to crank the preamp volume while listening. Am I wrong? If so, why?

Thanks!
If you love Hi-REZ TAPE HISS, you're REALLY going to love Stereo Central

User avatar
lukpac
Top Dog and Sellout
Posts: 4591
Joined: Wed Apr 02, 2003 11:51 pm
Location: Madison, WI
Contact:

Re: Digital gain increase question

Postby lukpac » Mon Dec 26, 2005 7:06 pm

Crummy Old Label Avatar wrote:Say you have a digital sound file that peaks at -12.56 dB. Now, say you use your editing program to increase the gain +12.55 over the enitre file. The file now peaks at -0.01 dB.

My question is this: why do some consider this to be a very bad thing thing to do to a sound file? By making sure that the peak remains under 0 dB -- and assuming the WAV editor doesn't use screwy algorithms, I can't understand what the harm could possibly be. Yes, you will increase the noise floor, but you'd do that anyway by having to crank up the listening volume of such a quiet track to begin with.

Common sense tells me that it would be best to increase the gain in the digital domain first, rather than have to crank the preamp volume while listening. Am I wrong? If so, why?


I think it's only bad in the sense that you're doing digital processing that you don't really need to be doing. I honestly don't know if I could hear the difference or not...

Although, if you go up exactly 12 dB, depending on the software, you'll just shift things over two bits, rather than doing lots of complex math.
"I know because it is impossible for a tape to hold the compression levels of these treble boosted MFSL's like Something/Anything. The metal particulate on the tape would shatter and all you'd hear is distortion if even that." - VD

User avatar
Crummy Old Label Avatar
Posts: 1226
Joined: Wed Aug 25, 2004 5:55 pm
Location: Out of my fucking mind

Postby Crummy Old Label Avatar » Mon Dec 26, 2005 7:28 pm

Yeah, I figured that digital phobia was the reason that such an operation was considered to be a no-no. Of course I understand that the complex math needed to increase the gain can theoretically screw things up, but I just wonder how likely it really is that such potential errors would be audible?

The reason the question comes up is that I have a CD from 1991 which is just RIDICULOUSLY low in volume. I ripped it, put the sound files into Peak, and, lo and behold, most of the tracks peak at around the -21 dB mark! Someone must have been asleep at the switch when this thing was mastered. (The CD in question is the ESD release of Henry Cow's Unrest.)

What I did was increase the gain on each individual track, taking the peak of each one all the way up to -0.01. I've been listening and comparing, and I'll be damned if I can notice any artifacts. I would guess that in the scheme of things, a mere gain increase is a relatively straightforward task for the software to perform.

But I will take your advice and try increasing by whole integer dB steps and see how that audibly compares to the original and the -0.01 peak files. Thanks for weighing in.
If you love Hi-REZ TAPE HISS, you're REALLY going to love Stereo Central

User avatar
lukpac
Top Dog and Sellout
Posts: 4591
Joined: Wed Apr 02, 2003 11:51 pm
Location: Madison, WI
Contact:

Postby lukpac » Mon Dec 26, 2005 9:54 pm

Crummy Old Label Avatar wrote:Yeah, I figured that digital phobia was the reason that such an operation was considered to be a no-no. Of course I understand that the complex math needed to increase the gain can theoretically screw things up, but I just wonder how likely it really is that such potential errors would be audible?


Not really sure, although I generally err on the side of caution. Of course, it isn't like you couldn't just burn another copy later if you noticed issues...

An interesting semi-tangent: my A/D stage is a DI/O, and I had always kept the tube thingy on "warm". I never really noticed a difference between "warm" and "clean", other than a slight volume difference, but I kept it on "warm" anyway. One day I was switching between the source and the computer output, and noticed the computer output seemed a bit off - grainy. For some reason I tried changing it to "clean", and sure enough, the "grain" went away. What was really interesting is I only noticed the differences at *low* volumes - if I turned things up I couldn't hear a difference. Anyway...

But I will take your advice and try increasing by whole integer dB steps and see how that audibly compares to the original and the -0.01 peak files. Thanks for weighing in.


Keep in mind the steps would be 6 dB, not just integer. 6 dB is doubling the volume, which digitally is shifting things over one bit. If your software works that way, of course. Non 6-dB steps would result in total recalculation.

If that matters.
"I know because it is impossible for a tape to hold the compression levels of these treble boosted MFSL's like Something/Anything. The metal particulate on the tape would shatter and all you'd hear is distortion if even that." - VD

damianm
Posts: 75
Joined: Sat Nov 27, 2004 7:46 am

Postby damianm » Tue Dec 27, 2005 4:27 am

I don't know the first thing about digital audio, but from what little I've read, I've got to ask: wouldn't amplifying a low-amplitude waveform in the digital domain result in quantization noise (which is, I understand, the result of having used fewer bits to represent the signal because of the low-level A/D conversion) becoming more noticeable?

As I said I don't know the first thing about this and would appreciate it if someone called me up on any bs I might've just typed.

User avatar
Crummy Old Label Avatar
Posts: 1226
Joined: Wed Aug 25, 2004 5:55 pm
Location: Out of my fucking mind

Postby Crummy Old Label Avatar » Tue Dec 27, 2005 6:18 pm

Krabapple knows more about dither than anyone else here. I think he could answer that question. (Please do, Krab!)

OK, so theoretically, what would be the best way to increase the gain? Would it be:

1) Taking the original 16-bit file and converting it to a higher bitrate first? (And which bitrate? 24 bit? 32 bit? And if 32, which is better -- integer or floating point?) Or is this a waste of time?

2) Applying the increased gain on the high bitrate file. Then:

3) Dithering back down to 16-bit.

I don't doubt that applying all these algorithms to a sound file will introduce artifacts and (possible) errors; I would just like to know how audible any of it would be.
If you love Hi-REZ TAPE HISS, you're REALLY going to love Stereo Central

User avatar
Crummy Old Label Avatar
Posts: 1226
Joined: Wed Aug 25, 2004 5:55 pm
Location: Out of my fucking mind

Postby Crummy Old Label Avatar » Tue Dec 27, 2005 6:31 pm

Not that i really know what the hell I'm doing, but later tonight I'm going to take a a few of the original files and convert them to both 24 and 32-bit, then apply the gain increase. I'll also dither those files back down to 16-bit. The editing software I'm using, Bias Peak (not any other choices if you're using a Mac) includes the POW-r dither algorithm.

I just found this article about POW-r dither, written by Bob Katz. He recommends using the "POW-r 3" setting," so I guess I'll give that a try. Shouldn't take long to do, and it may be interesting to see if it sounds any different.

Again, I'm totally confused about whether it's a good idea or not to make a 16-bit file a higher bitrate before doing any of this --- or even which bitrate it should be, but what the hell. I'm curious, so let's see what happens.
If you love Hi-REZ TAPE HISS, you're REALLY going to love Stereo Central

User avatar
Xenu
Sellout
Posts: 2209
Joined: Mon Apr 07, 2003 8:15 pm

Postby Xenu » Tue Dec 27, 2005 10:34 pm

Crummy Old Label Avatar wrote:Krabapple knows more about dither than anyone else here. I think he could answer that question. (Please do, Krab!)

OK, so theoretically, what would be the best way to increase the gain? Would it be:

1) Taking the original 16-bit file and converting it to a higher bitrate first? (And which bitrate? 24 bit? 32 bit? And if 32, which is better -- integer or floating point?) Or is this a waste of time?

2) Applying the increased gain on the high bitrate file. Then:

3) Dithering back down to 16-bit.

I don't doubt that applying all these algorithms to a sound file will introduce artifacts and (possible) errors; I would just like to know how audible any of it would be.


I don't think at all, honestly, regardless of what should be the case. I've had two WAVs from different "sources" cancel out perfectly after volume matching, whereas you'd think that dither errors would prevent that from happening (e.g. the Japanese Victor Kinks discs, which have swapped channels *and* polarity *and* different volumes cancel out perfectly with the PRTs when both are normalized to 98%).
-------------
"Fuckin' Koreans" - Reno 911

User avatar
J_Partyka
Posts: 180
Joined: Mon Sep 22, 2003 6:56 am
Location: Pennsylvania

Postby J_Partyka » Wed Dec 28, 2005 6:39 am

Crummy Old Label Avatar wrote:Again, I'm totally confused about whether it's a good idea or not to make a 16-bit file a higher bitrate before doing any of this --- or even which bitrate it should be, but what the hell. I'm curious, so let's see what happens.


I'm wondering the same thing, so I'd definitely be interested in your results. One of the reasons this is an issue for me is that, as an iPod user who always listens in shuffle mode, I get really irritated by drastic level fluctuations between songs, and I've been trying to come up with a way to address this. "Smells Like Teen Spirit" from the MoFi Nevermind should not play more quietly than an acoustic song from the latest Ryan Adams album. (I do use the iVolume application, but this only makes the problem less severe; it doesn't eradicate it.)

User avatar
krabapple
Posts: 1615
Joined: Wed Apr 16, 2003 4:19 pm

Postby krabapple » Wed Dec 28, 2005 12:40 pm

Yes, it's good, safe practice to do any digital processing other than simple editing, in higher-bit domains. Cool Edit has an option for opening everything in 32 bit, for that reason. That way the wordlength errors that get introduced during digital processing are likely to have no significant audible effect when you dither back down to 16.

But for peak normalizing, the issue is not dither (assuming you follow the above practice) , it's digital overs. You could be introducing them even though everything looks like it's peaking under 0 dBFS. So, what's the safe level to peak at, given thet most of us don't have the sort of digital meters that detects such things? I wish I could tell you. Some recommend -6 dB, but I suspect that's overkill. Katz recommends -3 I think. I use replaygain (a psychoacoustics-based gain leveling DSP) on playback of everything these days, and it tends to reduce gain by at least a dB or two...often signficicantly more in modern recordings. I suspect that takes care of it for me.

Replaygain spec is related to POW-r3 and Katz recommendations, but not in a simple way.

Digital overs explained
http://www.cadenzarecording.com/papers/ ... ortion.pdf

on dither:
http://www.answers.com/topic/dither-1

on replaygain (possibly a bit out of date)
http://wiki.hydrogenaudio.org/index.php ... Replaygain
http://replaygain.hydrogenaudio.org/
"I recommend that you delete the Rancid Snakepit" - Grant

User avatar
lukpac
Top Dog and Sellout
Posts: 4591
Joined: Wed Apr 02, 2003 11:51 pm
Location: Madison, WI
Contact:

Postby lukpac » Wed Dec 28, 2005 1:04 pm

krabapple wrote:But for peak normalizing, the issue is not dither (assuming you follow the above practice) , it's digital overs. You could be introducing them even though everything looks like it's peaking under 0 dBFS. So, what's the safe level to peak at, given thet most of us don't have the sort of digital meters that detects such things? I wish I could tell you. Some recommend -6 dB, but I suspect that's overkill. Katz recommends -3 I think.


Interesting article, but unless the recording is so compressed that almost everything is right at 0dB, who cares? When I do transfers, for example, I might have a few spots here and there that come close, and if they clip a little, I really don't care.

An interesting thought:

http://www.rogernichols.com/EQ/EQ_2000_02.html
Don’t tell anyone this, but during the mastering we turned up the overall level of the whole album 6dB. It was exactly a shift up of 1 bit. No math would be performed on the data to raise the noise level, but because of it there were a few overs on loud passages during snare drum hits. We listened carefully and decided that since the overs were not audible that we would leave them alone.
"I know because it is impossible for a tape to hold the compression levels of these treble boosted MFSL's like Something/Anything. The metal particulate on the tape would shatter and all you'd hear is distortion if even that." - VD

User avatar
krabapple
Posts: 1615
Joined: Wed Apr 16, 2003 4:19 pm

Postby krabapple » Wed Dec 28, 2005 5:16 pm

The problem is that the peaks need not be right at 0 dB for them to register as overs in the playback DAC. IIIRC in the article, it's saying that peak levels *apparently* as low as -6 dBFS can still register overs.


"Yes, there are a lot of problems with converters and headroom, but it is not because of the filters and the way in which they dissipate out the transition band frequencies. It is the fact that the signal itself exceeds 0dB regularly even when the samples do not." -- Aldrich


As to the audibility of digital clipping, it's true that not all clipping matters; some say that it requires 10-28 consecutive samples to be audible; others note that density of clipping matters too (lots of shorter clipping but spaced closely together).

Still, why clip at all if you don't have to?

Nichols' article on jitter gives me a pain, because he's usually a good source. Nothing verified in that article, though. Just the usual 'I am a recording engineer, I heard it, it is true' sort of claim.


Here btw is Katz' own article on digital overs
http://www.digido.com/portal/pmodule_id ... age_id=36/
"I recommend that you delete the Rancid Snakepit" - Grant

User avatar
lukpac
Top Dog and Sellout
Posts: 4591
Joined: Wed Apr 02, 2003 11:51 pm
Location: Madison, WI
Contact:

Postby lukpac » Wed Dec 28, 2005 5:28 pm

krabapple wrote:The problem is that the peaks need not be right at 0 dB for them to register as overs in the playback DAC. IIIRC in the article, it's saying that peak levels *apparently* as low as -6 dBFS can still register overs.


Right...I'm just not sure why it matters.

Still, why clip at all if you don't have to?


Why waste bit depth/amplitude if you don't have to?

Nichols' article on jitter gives me a pain, because he's usually a good source. Nothing verified in that article, though. Just the usual 'I am a recording engineer, I heard it, it is true' sort of claim.


Blind tests of some sort would have been nice, as would a reasoning for all of this, but it certainly describes what a lot of people claim to hear. I give him kudos for 1) saying it *is* dependent on the player, and 2) saying you can copy a "bad" disc and "properly" burn it and it will be "good".

If only VD would listen.
"I know because it is impossible for a tape to hold the compression levels of these treble boosted MFSL's like Something/Anything. The metal particulate on the tape would shatter and all you'd hear is distortion if even that." - VD

User avatar
krabapple
Posts: 1615
Joined: Wed Apr 16, 2003 4:19 pm

Postby krabapple » Wed Dec 28, 2005 5:32 pm

Why it matters

http://www.audioholics.com/techtips/spe ... ipping.php

and see the link therein to one of the original AES papers on the topic...from which I draw this quote

All of the domestic CD players investigated have shown difficulty dealing with 0dBFS+ levels that can
easily occur on modern CDs. New models are actually worse than older types relying less on
oversampling and more on analog filters.
We have not investigated how seriously audio quality is subjectively affected, nor have we made any
listening fatigue tests concerning 0dBFS+ levels. However, modern CDs contain these kind of signals and
modern CD players are not designed to reproduce them without distortion.
There appears to be plenty of reasons for concern about the quality of audio when hot mastering levels are
to be reproduced at the end listener.
To make things worse, the mastering engineer is neither able to hear nor see when the level danger-zone is
reached.



The conditions of audibility of digital clipping is not a settled issue...which would make it matter to me. Moreover, you are not wasting headroom if you are actually *using* it -- and the crux of the digital overs articles is that you may be using it when you don't realize you are. If you had a miscalibrated 'VU' meter on your cassette deck, such that you were going into the red without realizing it, and then you fixed it, would you call that 'wasting headroom'?
"I recommend that you delete the Rancid Snakepit" - Grant

User avatar
lukpac
Top Dog and Sellout
Posts: 4591
Joined: Wed Apr 02, 2003 11:51 pm
Location: Madison, WI
Contact:

Postby lukpac » Wed Dec 28, 2005 6:04 pm

krabapple wrote:Moreover, you are not wasting headroom if you are actually *using* it -- and the crux of the digital overs articles is that you may be using it when you don't realize you are. If you had a miscalibrated 'VU' meter on your cassette deck, such that you were going into the red without realizing it, and then you fixed it, would you call that 'wasting headroom'?


If it sounded fine and then I adjusted my levels down because the meters changed, yes.

Look, if your source is already so compressed that the peaks aren't very, well, "peaky", then yeah, I'd say you should be careful. But if your peaks are quite a bit louder than everything else, and they are usually things like drum hits or cymbal crashes, why record at a lower level if you don't have to? I suppose it isn't *quite* as important with digital as it is with analog (tape hiss), but there is still a lower limit to what you can record, and it seems to me that you'd want to stay as far away from that as you can, without affecting the other end.

If recording right at 0dB sounds fine now, why go down to -3dB or something if it might not even affect anything?
"I know because it is impossible for a tape to hold the compression levels of these treble boosted MFSL's like Something/Anything. The metal particulate on the tape would shatter and all you'd hear is distortion if even that." - VD