TECHNICAL question on loudness of CDs

it’s better to mix into compression already added to the 2-bus. pretty much standard practise in the industry and why the SSL 2bus compressor becuase famous/infamous.

If you add compression afterwards it’ll alter your mix. So better to mix into compression. Some mix into EQ as well. For example if you want a bright or a bassy mix then simply add it on the 2-bus, saves a lot of channel Eqing :slight_smile:

:slight_smile:

Nice, Karl, thanks. Thinking hard about it, lots closer to understanding it than before.

From what you wrote, I’m thinking that overshoot/undershoot occurs with every D/A conversion, not just the ones at the top of the scale … is that correct?

If so … what is it that happens at the top of the scale that makes it more “unmusical” than overshooting at lower target voltages? For example, if the D/A happened to put out 1.001V where 1V was the target, why does that have so much worse a rep than an overshoot to 0.9001V where 0.9 was the target?

If the overshoot occurs below 0dbfs, the top doesn’t get flattened. I have a background as an electronic technician but am not an engineer so I have a general understanding of the concepts but I’m certainly not an expert.

I’ve seen o’scope traces of amp’s that have been “railed” to max output (pushed to where the output voltage is equal to the power supply rails) and when that happens you can see lot’s of electrical ringing in the signal. In audio terms this would be lots of noise/distortion/harmonics etc…

Thats my understanding.

Again, I’m not an electrical engineer but my best understanding is that, at the top of the scale you begin to get kind of a peak limiting effect…a flattening of the peaks of the waveform. This flattening is similar to the "railing"effect I described above. It generates lot’s of additional harmonics (distortion).

If you want a really good explanation you should pick up a copy of Bob Katz’s “Mastering Audio”. He goes into great detail and touches on lots of additional factors (various filtering mechanisms etc… that are employed in D/A convertors and how these react/interact with overdriving signals.

Then by definition it isn’t an “overshoot”

With well designed DA converters inter sample peaks should not be a problem, the problem comes from less well designed DAs that may have increased distortion when the D side hits or is very near 0dBFS.

Its not so much the slew-rate overshoot, but a function of the DA smoothing, at the top of a peak where there may be two or more adjacent virtually or actual full scale samples, in order to recreate the correct wave shape the best fit curve will need to rise above the containing samples, thus an inter sample peak occurs.

It definitely is a problem with ‘consumer’ gear rather than quality gear from my experience.

Hey Dave S - sorry, are you saying the overshoot is just a problem with “consumer” gear, or are you saying that “quality” gear overshoots also, but the auditory results are better?

Is there a way to examine waveforms and see whether the A/D converter “overshoots” or not? How would one even know if there is an “intersample peak”? To my current way of looking at it, it would be hard to know if two samples right at 0dB had an “intersample peak” between them, or if that is actually what was recorded?

Thanks -

Inter sample peaks are a function of D to A conversion and occur at the playback converter.

From my experience it can be a problem with cheap converters. For example on board sound chips, cheap cd players, some mp3 players etc.

What I did was insert Stillwells ‘Bitter’ plugin in the cubase master bus.

Hope that helps

Thanks Dave, yah, it does. Though I wouldn’t call my A/D-D/A converters high class (see specs below), I’m at least reasonably confident that they’re better than a Soundblaster sound card, cheap CD player, etc. :smiley: :question:

Alexis…

I think this thread has strayed well into the extreme esoteric and away from the point of your original post (how to maximize playback level).

I’ll offer you the gist of my understanding re…intersample peaks, digital clipping distortion etc…

  1. Lot’s of complaints in the early days of digital about crap sound quality if digital was overloaded (ie…>0dbfs). Early converters were not nearly as good as todays (which are still not nearly perfect). Subsequent studies came to a view that digital distortion sounds worse than analog because it tends to be odd harmonics whereas analog tends to generate more even harmonics.

Ture/not true??? I couldn’t tell you on the basis of having personally SEEN the scope traces proving it. I’m pretty certain that I’ve heard it myself and am reasonably convinced that it doesn’t sound good. Record at 24 bit’s and keep the levels down well below 0db

  1. Maximizing playback levels…I certainly dont know all (or even most) of the tricks. this is evidenced by the fact that I’m always wondering why I cant get my own mixes as hot as those on CD’s and radio. I get better with each try and find little things but I have no silver bullet to offer.

Maximizing playback levels…

Careful EQ’ing is my main approach. Most stuff can be eq’d to where it sounds a bit thin on it’s own. Doing this eliminates alot of sonic energy thats un-needed allowing the mix as whole to be turned up a touch higher. Try this, record a tune with drums, bass and piano. Do no eq on them, just mix them to where they sound pretty good. Bring up the stereo bus/track faders until the peaks shown on the main buss reach -0.2 (ie, you know you’re not overdriving but you’r very close to 0db).

Now take the piano track, turn on eq1 (the HPF/shelf eq. Turn down the gain as far as it’s go, adj the Q to get rid of the hump on the curve (in other words, you should see a deep rolloff at the low end rising to 0 with no hump). Now increase the freq on that eq up to around 250-300 hz. You’ll definately hear the bottom end being carved out of the piano but if you un-solo it, the bass and drums and will fill much of that in. Do this and now look at what your stereo bus peaks are.

They’ve probably fallen a good bit below the -0.2 that you had to start with. You can now turn up the track a bit to get the peaks back up to -0.2… you’ll get more perceived volume but still be at the same peak level as before.

You can do this on all the instruments, use EQ to carve out the un-needed frequencies. This reduces the amount of sound (in this case, electrical) energy, reduces the peaks, leaves more space in the mix for other instruments etc…

Lastly, I must admit, I do like plugging something like a Waves L2 or L3 maximizer on the stereo bus. Almost without fail, I find I can compress the very top peaks, have no negative effect on the sound of the song and no real perceptible effect on the dynamics even though I know it MUST be reducing them. But in so doing, I get a significant increase in the playback level without having to push the peaks right up to 0db (ie…I set the peak limiter to -0.2 or so to give enough room for inter-sample peaks etc…).

I’m sure there are other tricks that ME’s use (multiband compression etc…), but I dont know them, or dont know them well enough to explain them.

All the best,

Karl

Thank you, Karl that was really helpful, and I will try that (will have to wait till next week, I’m away from the music for the weekend :frowning: ). Great explanation, you’ve got a future in teaching if you want it, I believe.

I went to your website, listened to some of your tunes on “Made with Steinberg”, looked at your equipment - great tunes, awesome voice, professional production, and awesome looking set up! I noticed how your audio monitors are out to the side, so as not to get reflections off your desk - nice touch. Gotta ask, too - how big are those computer monitors?

Your tunes clearly DO have dynamics, but I notice their waveforms are much more uniform than mine - my soft bits are WAY softer “looking” than yours, and your tracks in general don’t have the centipede “fur spikes” sticking out like mine do. I’ll try the things you suggested, in addition to manually clipping peaks, and read up a little on the Waves maximizers also.

Thanks again!

I think the discussion on intersample peaks and D/A converter distortion is leading you down the wrong path. Just a couple of points from what I understand:

  1. Commercial recordings peak at 0db because they use tools that include normalizing the final result to 0db. You can also use them. You could achieve it by playing your track in Cubase, then look at the master fader, see what the peak is, and raise the fader that amount. That’s all normalizing does. But you don’t want to do this until you’ve already done whatever mastering processing is done, so if you’re sending out, you want to peak at -3db so mastering has room for processing.
  2. Apparent loudness is not achieved through peak volume, as has been pointed out – it’s the average loudness which is the RMS loudness expressed in a decibel range representing the dynamic range of the source. All other things being equal, normalizing doesn’t change the RMS loudness.
  3. A smaller decibel range (louder sounding) is arrived at in lots of ways. You could put a limiter on any tracks that have a lot of transients to reduce them. You could put a limiter on the master. Compressing reduces the dynamic range. Loudness plugins use combinations of limiting, compressing, and normalizing to do these things all in one go. If you overdo it with these tools, you are Metallica.
  4. If your final output sounds distorted, it’s probably not anything to do with intersample peaks or D/A converters (although I’m willing to believe it’s possible if Dave S says he’s had it). My point is that there are so many other easier ways to introduce distortion in your output, and you should look for those. But of course that wasn’t your original topic of the post.
  5. It must not be that bad to peak at 0db, since all your favorite CD’s do that and you didn’t toss them out for it!

I’ve heard this from several sources over the years, but I’ve yet to hear of a mastering house that isn’t capable of lowering the the level of something if it’s too high for them.

I used to use maximisers until I tried simply clipping off the short peaks and found it sounded much better. Problem with maximisers is that they distort the whole track… When you clip the peak you only distort at the point the short peaks that are clipped off.

Of course the effect of distortion is subjective and some people may prefer the distorted maximiser version to the cleaner clipped version. :slight_smile:

I remember you wrote something like that some years ago, which is why I did what I did as described at the top of this thread. I found out it very time-intensive!

Paul, why would a maximizer “distort the whole track” … as opposed to just the parts it is clipping, which (theoretically??) would be just what you would do manually?

Well, that kind of depends on the quality of the plugins you use…

A maximiser doesn’t just clip though. They are a form of intense limiting which works with attack and decay times on the signal.

And sure a more quality maximiser would probably distort the whole track less or in a more pleasing way, but it will still distort it.

You can try it by maximising a track and then performing a cancel test with the original ( volume adjusting to get max cancellation ).

The cancel test with a track that’s simply had it’s volume raised to clip off the peaks will only show residue in the cancel output where the peaks have been clipped off.

Anyway … maximisers versus clipping. Whatever sounds best to you. :slight_smile: - I prefer clipping

Great experiment! After thinking about it some more … is the lack of cancellation simply because the audio has been run through some extra circuitry (which is never “clean”), or is it something specific to a limiter (as opposed to, say, an EQ)?

Thanks -

Well we’re talking in the digital domain, so there shouldn’t be any differences from a simple unaltered digital signal. So unless the digial maximiser is also adding some analog emulation then any artifacts introduced aside from the peaks is a manifesting of the limiting process. I remember the Waves maximiser having decay times, so that’s one definite contender.