Subtitles prepared by human
Hi, and welcome back. Apologies for the slightly click-baity title. But yes, there really is a secret of maximum loudness, and it works regardless of the genre, the medium or the platform. Obviously I'm not going to give it up yet, you'll have to keep watching, because we need some background to properly understand it. I remember when I first got my first little stereo system as a child, thinking that the volume control set an absolute loudness for playback. But later on, when I started using my dad's hi fi to copy vinyl records onto cassette, I discovered that the record level could make a big difference: set it too low and my copy would be really quiet and buried in noise; but set it too high and I'd get horrible distortion. With some material I noticed that pushing the levels a little into the red actually made it sound better… but that's not the subject of todays video. The point is, while you had a little leeway to choose how hot you printed to tape, the nature of the medium dictated that most material sat around the sweet spot where you get the best signal to noise ratio with the least distortion.
When digital audio was invented, it was originally assumed that people would continue to work the same way: the nominal unity level of analogue recording would be equivalent to -18 or -20 dBFS, with the extra dynamic range reserved for headroom. But of course with digital audio, that headroom is just as clean and linear as the rest of its dynamic range. There's no distortion penalty for pushing the levels right up to full scale, you just get a louder master. So that's what people did. And then people discovered that, actually you could push the levels a little bit further and deliberately clip the converters, and that this clipping would be inaudible if it were brief enough, or might even sound subjectively good. And then the first digital brick wall limiters came along, and the loudness war really got going in earnest. Now you could push those levels harder… and harder… and it wouldn't clip, it would just get louder… and suddenly everyone was trying to be louder than everyone else. Let's pause here to discuss the elephant in the room:
yes, louder is better. At least until you approach the pain threshold anyway. Below that, every dB of extra loudness will translate to perceived extra clarity, extra solid bass, more detail, more depth, more width. Whatever emotional impact the music has at a low level, it will have more of it when turned up louder. But real loudness is not the same as artificial loudness. Real loudness takes the whole waveform and scales it up. Assuming you're not damaging your hearing, this is a good thing. Real loudness is why a live gig can be so exciting: the natural, uncompressed dynamics of a drum kit, augmented by a good PA system, provides a visceral experience that you'll never recreate with a hi fi system and a studio album. Artificial loudness, as created by a brick wall limiter, is not the same. Increasing the average levels makes the audio seem louder, but the peaks are now smashed down into the body of the mix, and this has consequences. Not surprisingly this can reduce the impact of your transients,
making them less punchy or snappy. But there are also more subtle side effects, such as a loss of depth and space: while real loudness can fill a room and immerse you in the music, artificial loudness forms an oppressive wall that pushes you back in your seat. Perhaps most egregiously, reducing the peak to average ratio can suck the life out of a mix, making it flatter, more boring, less exciting… if it's an upbeat energetic rock song this is perhaps the worst thing you can do: the listener won't know why, but the music simply won't be as exciting as a classic rock mix with the dynamics intact. Of course, at this point I'm obliged to point out that we've come a long way since the first brick wall limiters were released, and limiters have become much more sophisticated and powerful. Pro-L2 with its Modern limiting style can smash a mix really hard, yet still retain a remarkable sense of punch and definition. But no amount of clever program dependency can change the fact that you're reducing
the relative size of the peaks, and filling in the gaps between them. A clever person once said that music is the space between the notes… removing or reducing the space between your transients risks sucking some of the musicality out of your mix. However: when used carefully, a limiter can make the audio significantly louder with no negative impacts at all. And usually, significantly louder still with only very minor side effects. So this leaves us with a dilemma: if we don't use any limiting at all, our mixes will play back much quieter than the majority of other releases. Of course your listener could just turn it up… hold that thought because we'll come back to it… but you'll worry, quite reasonably, that they won't, and your mix will just end up sounding weaker than its peers. But on the other hand, competing with the loudness levels of many modern releases is also going to make your mix sound weaker than it potentially could. So, what's the optimum loudness to aim for when mastering? Or to put it another way, at what point
is the benefit of the extra loudness outweighed by the disadvantage of a low peak to average ratio? Well, we need some way to quantify loudness in order to answer that question, so I'm going to have to talk about metering; how we measure the loudness of a signal, which is more complicated than you might initially expect. Let's start with the channel meters in a DAW. These are quite simple: they flip the negative half of the cycle to positive, then display the sample values on a decibel scale. Of course the signal is continually oscillating through zero, and the meter would be unreadable if it showed that, so the values shown decay slowly to smooth out the dips, and display this sine wave as a constant level. Hence, if the sine wave stops suddenly… the meter nevertheless decays gradually. This is perfect when you're tracking vocals and you need to know how close you are to clipping. But it doesn't cut the mustard for mastering, for a couple of reasons. The first problem is that this meter just reads maximum sample values. But when this signal is converted to analogue the original curvy analogue signal will be perfectly recreated,
and its normal for the peaks of the analogue signal to be in between samples, and higher in level than the sample values either side. Measuring the actual peaks of the waveform requires a more sophisticated true peak meter, that calculates values in between samples. More on this later. Perhaps more significantly however, peak metering… and also true peak metering… tells us almost nothing about how loud the signal will sound. This is partly due to the nature of human hearing. Its not flat in frequency response, so a sine wave at 2k… will seem much louder than a sine wave at 200Hz… even though both have the same peak levels. Also, a long burst of 2KHz… will seem much louder and more obnoxious than a very short blip… even though both are again at the same peak level. A more real world example would be drums and distorted electric guitar: if I normalise these to the same peak level, the guitar completely obliterates the drums…
the drums need to peak at a significantly higher level than the guitar for the two to sound balanced, because of the short spiky transients in the drum track. Finally, peak levels don't really mean much anyway! The easiest way to demonstrate this is with a sawtooth wave: here's how that looks on Pro-Q3's analyser, with a low fundamental plus a harmonic series above it. Each of these harmonics is essentially a separate sine wave. Now I'll add a high pass filter: I’ll start with it set way too low to have any audible effect on the sawtooth wave… but notice, the peak level is already reading 2dB higher than it was. Now lets start to wind up the filter cutoff, until it starts to shave away the level of that lowest partial… I’m sure you’ll agree that this doesn’t sound louder than the unfiltered version… and yet the peak level of the filtered version is a full 6dB higher than the unfiltered version. If you've ever wondered why an EQ cut or subtractive filter can make your mix apparently get louder, this is what's actually going on: phase shift has caused
all the individual partials to add up differently, and the peaks of the wave have got higher as a result. But it's not actually louder in any real sense. So measuring the peak levels is pretty much useless when trying to determine how loud something will sound. That needs a different approach, and there have been a few over the years: the classic VU meter relies on the physical ballistics and inertia of the needle to smooth out the peaks, and display more of an average level instead. This is much better than a simple peak meter. But you need a little practise to interpret what a VU meter is telling you, especially when it comes to signals with lots of low frequency content, which tend to pin the meters even at perfectly sensible levels. Another approach is to measure RMS levels, or Root Mean Squared: without delving too deeply into the maths involved, this averages out the levels over time; “mean” being simply a more mathy way of saying average. This solves the problem of phase shift: inaudible changes in phase will not cause changes in RMS levels, and our sawtooth wave doesn't read higher when I high pass filter it.
It also potentially solves the problem of duration, as we’re averaging the signal levels over time. But the question is, how much time? If you average levels over a short 50ms window you’ll get a much faster moving, bouncier reading than if you average over 500ms… or 2 whole seconds… With longer, larger windows RMS levels correlate quite well with perceived loudness. But we still have the same problem as VU meters with bass heavy signals that tend to read too high. So some meters use weighting to try to emulate the non flat frequency response of human hearing. Basically this means a filter to boost the upper midrange and cut the low and high extremes, so the meter responds more or less respectively. This is a standard A weighting curve as used for measuring environmental noise levels… I told you measuring loudness was more complicated than it seems! But we do have a better option these days, thanks to the broadcast industry which has developed a new standard way to measure loudness,
with a new unit of measurement known as Loudness Units or LU for short. This new scale is based on RMS style averaging, but includes frequency weighting to avoid the problem of over reading bass heavy signals, and also defines the averaging window used. In fact it defines three different ones, called Momentary, Short Term and Integrated. Momentary averages signals over 400ms, and gives us this relatively bouncy display labelled M in Pro-L2. Short Term uses a six second window, and provides a much steadier reading that doesn't move much in response to drum transients. This is the most useful measure of "how loud is the signal at this point in the song". The integrated reading is a little different: this averages out the levels over the whole song… or the whole album, or whatever you want to measure. In Pro-L2 you can click to reset before the start of the song… and by the end of the song, or the end of the album, you'll have a single integrated loudness value for the whole thing.
Your DAW or audio editor might also allow you to analyse clips offline to calculate an integrated loudness for the whole clip… This might seem like a perfect way to quantify the loudness of a song, but it does have one flaw which I can illustrate with Led Zeppelin: here's Immigrant Song… I won't play it to avoid copyright strikes, but I'm sure you know how it sounds: it kicks in hard, and rocks hard all the way through. Now compare this with Stairway to Heaven: this starts off famously quietly, and builds gradually all the way through. As a result this mix has a lower integrated loudness than Immigrant Song, even though the climax at the end is about the same. So actually the best way to derive a single loudness value for a song might be to isolate the loudest section, typically the last chorus, and measure the integrated loudness of just that section. Failing that, watching the Short Term reading during that section will do just as well. Note that Loudness Units correspond to decibels: if a mix has a loudness of -16 LUFS, and then you turn it up by 3dB,
it will now have a loudness of -13 LUFS, which helps to keep things nice and simple. So, now we have a way to measure loudness, you doubtless want some numbers: what loudness readings should you be aiming for? Should you even have a target level? And does it need to be different for different platforms or formats? Well, we'll get to that in part two, along with the secret in the title of course. Thanks for watching.
Watch, read, educate! © 2021