Can a track be recorded in TOO quietly?

[rasputin1963]

Something I've been meaning to ask you guys. I just recorded to audio track an organ riff. I see now that the track's volume peaks around -24dB. Is this too quiet for a track to be recorded at? Are there problems waiting down the line if a digital track is recorded in that quietly? Is there a "sweet spot" in the dB ladder where a musical voice is best recorded in?

 

Another question: Do tracks suffer in quality if you simply raise their gain, or normalize them, say, to have peaks at -3dB ? Is the digital domain more forgiving in this department than analogue used to be?

 

This question may seem trivial, but it's very much part of my learning at present.

 

Thanks, ras

 

[UstadKhanAli]

See if this helps.

[rasputin1963]

Thanks, Ken. EXACTLY what I need to study. And I learned an excellent new word: "gain staging". When the chela is ready, the guru will be there.

[blue2blue]

And now that you've gone and studied, I'll just note that a lot of folks use a mental 'red line' of -18 dB FS -- a target they have 18 dB of headroom on. Of course, this presumes 24 bit signal capture, kind of a foregone conclusion, but worth noting. It's 24 bits ~140 dB dynamic range that allows that freedom.

 

 

By the way, I don't know if you've seen this, David, but it's really good. It does a great job of explaining the digital side of the dynamic range issue, not to mention frequency/sample rates. (Or did it get posted in one of your threads elsewhere? I seem to recall posting it somewhere a couple times recently.)

 

[video=youtube;cIQ9IXSUzuM]

 

More vids, etc: http://xiph.org/video/vid2.shtml

[rasputin1963]

Wow, great video, Blue. Dat guy is smart, huh? I'm especially glad he went into the nature of Dither, for now I see what a "shaped" dither is. SONAR offers you a number of dither algorithms, and I never know which one to choose... as I can't hear any audible differences between them all. My brain is not as mathematical as his is, but it surely behooves me to know all this stuff.

[MikeRivers]

Something I've been meaning to ask you guys. I just recorded to audio track an organ riff. I see now that the track's volume peaks around -24dB. Is this too quiet for a track to be recorded at?

 

Well, technically, yes, but only if you're recording with a 4-bit system. With modern 24-bit systems, it's not a problem with recording.

 

Are there problems waiting down the line if a digital track is recorded in that quietly? Is there a "sweet spot" in the dB ladder where a musical voice is best recorded in?

 

Another question: Do tracks suffer in quality if you simply raise their gain, or normalize them, say, to have peaks at -3dB ? Is the digital domain more forgiving in this department than analogue used to be?

 

There really isn't a digital sweet spot. Older A/D converters tended to be less linear within a couple of dB below full scale, but modern converters work pretty well in this respect. It's a good idea to leave a few dB of headroom so that an EQ boost when mixing doesn't drive some frequencies into clipping.

 

In theory, you can boost the level digitally without adding noise or distortion, but you have to consider the signal-to-noise ratio of what's going into the A-to-D converter. If you have the preamp gaim cranked all the way up and that bird outside your window is still only peaking the record level meter at -30 dBFS, remember that when you boost the level of the track, you'll be boosting that preamp hiss along with the bird.

 

[blue2blue]

Recording at -24 means you've lost about 4 bits of dynamic range. So if you had a 16-bit audio engine as was happening in the days of Sound Tools), you would be down to 12 bits so it was a good idea to keep the levels high. However, these days with floating-point 32-bit audio engines (and since you're using SONAR, 64-bit double-precision), a few bits don't matter and with converters and interfaces that have 110 dB or so of real dynamic range, losing 24 dB still gives you 86 dB of dynamic range, which is 10 dB shy of the theoretical (in other words, good luck!) maximum for a CD.

 

So, bottom line is, don't worry about it. Personally, I like to hit around -6 to -12 dB because I'm not interfacing with any analog gear further down the line and that way I lose only a couple bits of resolution.

 

Finally, remember there are the "theoretical" and "real" bits of resolution. For example 16 bits of resolution gives you around 96 dB of dynamic range. However, circuit board layouts, noise, stray RF, component tolerances, etc. will bring that down considerably. That's why you need 20-bit converters to get 16 "real" bits of dynamic range. And while 24-bit converters theoretically can give 144 dB of dynamic range, figures in the low hundreds are typical in the real world.

 

That vid seems to hit the right spot for a wide range of people and tends to use examples and demonstrations that communicate concepts quite well. It was a big hit at a certain other gear oriented site where there are a lot of knowledgeable people, but a also a lot of newbs -- and a stubborn population of Dunning-Kruger exemplars. I think a lot of folks got a good bit of primary or remedial education via that vid. (Xiph and Monty Montgomery have some other explainer vids as well.)

[blue2blue]

One last tidbit on near-0 dB FS signals -- sample values are 'legal' up to 0 dB FS -- but the loudest sample value does not (necessarily) represent the loudest level in the reconstructed signal (and, if you think about it, is unlikely to) -- so, unless your playback DAC can accommodate these over-0 dB FS signals, you could end up with overload distortion in the output, even with technically 'legal' signals. For this reason, most decent DAC's have accommodation for a few dB of headroom above 0 dB FS in their analog output circuits. But cheap devices (and there are ever more and more cheap digital devices) may not -- so it's become common to check for such 'oversample' signals using something like SSL's old, free X-ISM oversample aware buss meter VST/AU plugin. It's no longer available from SSL, apparently, but at least one download site prominent in search returns for "X-ISM meter" seems to have it. Caveat downloader.

 

Too bad SSL appear to have ditched it, since they had a really nice explainer on how intersample overs occur (which also appears to be gone) -- such 'overs' are all very 'natural' and implicit to the process -- but, as noted, the problem is when signals with 0 or near 0 dB FS hit cheap DAC units in cheap gear on playback, which may not have adequate overhead on the analog end for handling the signals.

[WRGKMC]

Sound wise you have a huge amount of headroom so low levels aren't a problem tone wise, but it can be an issue if you plan on mastering a recording to commercial loudness levels.

 

So long as your final mix comes in around -17db RMS (-12db average) your can master the mix to commercial levels with most plugins properly.

 

Anything hotter and you may wind up running out of headroom running Limiters, multiband etc. Too low and you may wind up working the mastering plugins harder then needed which can increase hiss levels, depending on the type of plugins you use.

 

For example, I use Waves L2 for my brickwall limiter. If the file comes in around -12db I can use a -4 to -8db threshold and nail the volume of any commercial recording. If I had to boost it 10db or more the results are too gainy and harsh using that plugin. If the recording is too hot, I may not be able to bring the threshold down at all without clipping the hell out of the peaks.

 

So the problem isn't so much the levels affecting the sound quality as it is in targeting acceptable ranges using plugins. Plugins do have wide ranges, but there are sweet spots that focus on signals that are within reason, (much like analog gear).

 

This short tutorial works like a champ for mastering. Getting a completed mix some where in the ball park (all tracks combined, not a single instrument) is key.

 

 

 

http://hdqtrz.com/Files/Har-Bal_Mastering_Process.pdf

[Phil O'Keefe]

And now that you've gone and studied, I'll just note that a lot of folks use a mental 'red line' of -18 dB FS -- a target they have 18 dB of headroom on. Of course, this presumes 24 bit signal capture, kind of a foregone conclusion, but worth noting. It's 24 bits ~140 dB dynamic range that allows that freedom.

 

 

I don't consider -18dBFS to be a red line, but it is generally the level I try to target when tracking. It gives me plenty of headroom for unexpected peaks while keeping me low enough to insure I don't get any "overs." And I do generally expect the peaks to exceed -18dBFS now and again... but I'm one of those "no overs - EVER - under ANY circumstances! " kind of guys.

[MikeRivers]

One last tidbit on near-0 dB FS signals -- sample values are 'legal' up to 0 dB FS -- but the loudest sample value does not (necessarily) represent the loudest level in the reconstructed signal (and' date=' if you think about it, is unlikely to) -- so, unless your playback DAC can accommodate these over-0 dB FS signals, you could end up with overload distortion in the output, even with technically 'legal' signals. [/quote']

 

I was going to mention what seems to have acquired the name "intersample overload" or "true peak level," but since the question was about recording at too low a level, I passed on that, It used to not be a problem because, at least in the early days before the loudness wars started, CD pressing plants would reject a master that had any full scale samples, or at least warn the customer about what they considered a problem. Eventually, projects came in mastered with nearly as many full scale samples as there were cycles, so they just dropped the needle and let it fly.

 

It was only when people started complaining about distortion in commercialy released music that the peaks between two samples could be the cause. But we knew all along that it was from the heavy limiting which, by definition, is distortion.

 

It gives the audiophiles something else to be improved upon, and overall that's a good thing.