Following some discussions on twitter about mastering and stuff, I decided to run the current mix for Erlanger Programm through the EBU R128 metering, and here’s what came up:
For those of you not familiar with this concept: the European Broadcasting Union has defined a recommendation for programme levels on stations (of whichever kind), along with a measurement standard. What can we read from that?
First of all, the pseudo-unit “LU” is defined as -23dB related to digital full-scale (or LUFS). Then, this standard defines a kind of “average average” (which is the “Integrated” value). Moreover, there’s also gating (leading to very quiet passages not being considered), and finally, EBU recommends that things are not played louder than at 0dB LU with a tolerance of +/-0.5dB.
This is, I might say, very low by today’s standards, but it also considers the fact that stations typically want to use algorithms to furthermore increase loudness. Interestingly, the original movie standard (albeit with a slightly different algorithm) had beeen -20dBFS, which together with AES17 (another norm) will land us at around the EBU recommendation. Coincidentally, it is also K-20 by Katz’s K System, and with the magic of +6dB, it would then land us at 6dBLU (including AES17) for a nice pop/rock record.
Furthermore, the loudness standard set by iTunes radio around version 11 started some programme level pracices that would result in a level of 6.5dBLU (or roughly -2.5 in a K-14 world).
So what do those values above tell us? First of all, the true peak value is at +1.4dBFS, which is a lot, both relating to the R128 spec of that not to exceed -1.0dB, or that of typical mastering engineers which aim for a maximum for either true or digital peak in between -0.5 and -0.0dB.
True Peak Explained
To the side, you see an artificially (in the DAW) created sine wave with a peak level of 0dBFS and a frequency of 16kHz, in a 48kHz audio file. What can you see? The peaks don’t really get up to the top – in fact, if we analyze the digital peak, we get one of -1.25dB.
However, this does not change the fact that, by signal theory, this waveform extends to 0dB after it passes the LP filter in the DA conversion.
The problem? In a typical “get every ounce of loudness out” approach, an engineer might normalize this file to 0dB, resulting in the peaks being at over 0dB (meaning: in clipping), even though no single sample value is over 0dB.
This typically happens with very high frequencies – if you move close to Nyquist with 23900Hz, you’ll have a true peak that’s more than 3.7dB above the digital peak! Of course, those frequencies are rare as sine waves, but they do happen as transients – and those are typically the signals punching into the last few dBs of headroom.
Of course, my approach is typically to be safe, but not to be double-safe, meaning: it’s perfectly clear that you won’t get a problem if you keep your true peak below 0dB (unless the converter’s manufacturer, or someone else in the signal chain, fucked up). On the other hand, the true peak overshooting by several dB are rather theoretical cases. And finally, with the definition originally established by Sony that “everything below three samples in a row at full scale is not clipping”, the audio world had been pretty happy, unless using a DAC from the early 80s.
And finally, I have started to use a true peak limiter as the last element in the chain, and setting that to -0.1dB did result in this project’s test renderings ending up at a digital peak of -0.11dB – and that’s where I feel safe from both sides.
The Integrated Value
Next up: the integrated value (I) from above: this is essentially an integrated version of the short-time-averaged version of the input signal: it’s the average of the short-time average over the whole measurement time (here: the whole album), but…
R128 also defines a gating rule, or rather, references one. The idea here is to keep passages where the next song is announced etc. from disturbing the picture. On the other hand, very quiet passages (in relation to the average) will be discarded from the integrated measurement – which makes sense, not only for broadcast applications, but also for mastering an album, as a “forte” passage is typically the best reference to normalize to (or from).
That being said: the average (not counting in the super-quiet stuff) of that album is -4.4dBLU, or -27.4dBLUFS. That’s rather quiet, but not surprising considering the large amount of quiet sections on this album.
Looking at this from the perspective of a potential mastering engineer: I mentioned that iTunes target of 6.5dBLU which, in comparison the the value above, is about 11dB hotter. Which is just as well, because it gives the mastering engineer enough room to play with, and then some.
Momentary and Short-Time Values
In addition to the integrated value, which is averaged over the whole piece (in this case, close to 44 minutes), there’s the short-time (S) value, averaged over 3s, and the momentary (M) value, averaged over 400ms.
While all of these are displayed in a meter, Cubase provides readings for short-time and integrated in number format, as well as short-time and momentary peak values.
So the current short-term is not available (no wonder, as this pic was taken the moment the album was over), but the maximum was 4.7dBLU. The momentary maximum was at 8.2dBLU. Moving both to the LUFS domain, it means that the maximum momentary RMS was around -15dBLUFS.
More interesting from a programme standpoint is the S value of around -18dBLUFS.
The K-System World
Bob Katz’s K-System differs from the R128 approach insofar as it does not measure the “relevant average” of a programme, but is rather used to measure a good reference to a forte passage – that region where important statements in classical and most stuff in pop/rock happen, and which (in an interesting arrangement) is only superseded by some fortissimo passages which add another 5-6dB.
So for this, neither the I value from above nor the M or S maximums are a good reference, but can be well-used for comparison.
Instead, I decided to use the K meter (at the papa/K-20 setting) for a forte passage (the section right after the intro for part B). As you can see, we get something like -6dB, which is a tad bit quiet, even when later applying the AES17 recommendation and adding 3dB (something which is part of the K-System, but which I simply forgot to turn on when taking the pic).
On the other hand, from a mastering engineer’s standpoint, it totally makes sense: if you want the mix to come out at something like K-14, you get around 9dB to play with, and that makes the mastering engineer happy.
Sticking in the K-System world: I also measured the loudest passage of the album, and there we get a maximum of 2dB – which, in comparison to the -6dB, is slightly (namely 2dB) more than the aforementioned difference between forte and fortissimo. This, however, is something that a mastering engineer can gently work on (and maybe will have a better result than the mixing engineer confronted with the same task). Or you could simply consider this an especially big dynamic range – not only fortissimo, but fortissimo possibile.
The Mastering Mockup World
For a quick look into the future, I decided to have a simple comparison using the onboard offline tools of WaveLab6, and listen to the mastering mockup of this album – essentially, the version you’ve been listening to so far.
The average here comes out a -15.7dBFS. In that context, it makes sense to explain the minimal mastering chain:
- LP10 EQ with rather gentle settings (mainly taking out some LMFs and raising the HF).
- Limiter 6, with a gain of 14dB (!), programme limiter threshold of -4dB and peak of -2dB, peak limiter threshold of -1.5dB and true peak limiter peak of -0.1dB.
A first simple calculation would be to add the 14dB of gain to the I value from before and arrive at -13.4dBFS, or 2dB hotter then what we get here. There’s several possible explanations for this discrepancy:
- the EQ removed something in the LMF range, which usually is very crowded. Even if we’re only talking a dip of -1.5dB, this can easily reduce the programme level by 0.5dB.
- Driving a limiter by n dB will result in a programme level increase of <= n dB. The equality sign only applies if the limiter doesn’t do anything.
- Different kind of filters: both K-System and R128 use filters (C-weighted), which penalizes the very deep A I love (and use) so much by around 6dB.
So let’s just put a K-meter on the master, for the forte passage we’ve looked at before.
Switching to K-14 here, we’re arriving at +2dB. Adding the 14dB of gain to the -6dB we got before for the mix, and converting that to K-14, we get exactly +2dB – and find that we come out where we expect to come out.
A Brief Summary?
So we’ve looked at peak and average values, and did so for both a mix and for simple master. We furthermore looked at simple “RMS over everything” calculation, and at EBU R128 I values and M and S maxima (for the mix), and at K-System values (for both mix and simple master).
The end result, if there is any at this stage:
The mix, with a level of -14dB below a “proper master”, is quiet enough by all means. There’s no danger of getting too hot a mix to the mastering stage.
The supersimple master I have already is generally in the right ballpark. If the huge macrodynamic range will work for the finished album remains to be seen (or rather decided).