Professional Mastering - Analog & Digital
Professional Mastering
251K+ subscribers
I’ve been making these mastering, mixing, audio production videos for a while, and have come across a lot of cool concepts.
But despite all the research I’ve done for this channel over the years, I’ve only been surprised by something a handful of times.
This is especially true when something seemingly simple, something I do in just about every session, has very bizarre and unexpected implications for the sound quality.
I thought it would be fun to put all of these into 1 concise video, so let’s jump into
The Strangest Music Production Discoveries I Found from 7 Years of Research
Watch the video to learn more >
First up, we have...
So I was told about this from Fabien of Tokyo Dawn Labs - he’s their product and plugin developer, and Tokyo Dawn Labs releases a lot of fantastic plugins, many of which are free, so I’d recommend checking them out if you haven’t.
Not a sponsor, I just genuinely like them.
But anyway, he explained this to me, and I was so surprised by it I had to test it for myself.
In short, whenever we automate volume, a plugin parameter, panning, or anything that affects the amplitude of the signal, we clip the signal for the duration of the automation.
Seems fake, but it’s true. Here’s how and why that happens.
As you know, the sampling rate determines how many samples occur in a second. We often associate the sampling rate with the frequency response or the highest supported frequency in digital audio; however, it plays a big role in accurate amplitude quantization.
You can have a high bit depth, but if the sample rate is low, you’ll cause steps in the signal’s amplitude.
This isn’t a problem with any modern session, even at 44.1kHz. But, let’s say we affect the amplitude of a continuous waveform in a way that sums 2 of these samples, or 4, or maybe 8, what would happen?
We’d create the same stepped, inaccurate amplitude assignment as if the sampling rate has been reduced. Subsequently, the waveform isn’t a smooth, continuous transfer. But instead, a waveform with its amplitude chopped into blocks.
Whenever the signal’s amplitude is altered from its original form, the difference between the original amplitude and the incorrectly squared or stepped waveform is distortion. It’s the addition of new waveforms that is directly proportional to this difference.
So what does this have to do with automation?
As you probably guessed, automation causes the sample summing I mentioned earlier, by making changes to amplitude not on a sample-by-sample basis, but over the span of 6, 8, or however many more samples.
If I play a sine wave and automate the amplitude of the channel fader, you’ll notice large spikes of multiple harmonics.
If we freeze the display of this analyzer, we can observe that the artifacts from automation are pretty severe. You have to imagine that these artifacts become audible, or meaningfully interact with the original signal or other signals, when multiple tracks are automated at the same time.
And unfortunately, there’s no way around this unless the automated parameter affects the amplitude on a sample-by-sample basis. But everything I’ve tested, from EQ filters that change the amplitude, panning, which changes one channel’s amplitude, the output of a compressor, just about anything you can think of, introduces this distortion.
Fortunately, changes to the original recording, such as fades and clip gain, work on a sample-by-sample basis and don’t introduce this issue.
It’s an interesting and unexpected consequence of something that seems mundane, but gives us a good reason to use automation sparingly, and when possible, replace it with clip gain and fades.
Watch the video to learn more >
The definition of a limiter is incredibly broad - we could be talking about compressors with a ratio of 10 or 20 to 1. We could be talking about anything that limits the peak amplitude to a set ceiling.
But I think you know what I’m referring to when I say limiter.
It’s not a clipper that squares the waveform, it’s not a compressor with a high ratio, or a saturator that introduces soft clipping, etc.
It’s a plugin designed to keep the signal to a set ceiling without causing significant reshaping to the waveform.
Its purpose is to increase the amplitude without big changes to the timbre.
Some plugins that are labelled limiters or maximizers achieve this, but a surprising number introduce hard clipping.
For example, Izotope’s Maximizer does a pretty good job - if we observe the waveform, we’ll notice that a lot of settings result in a retained peak.
But, if I select IRC II, anything quicker than a value of 2 causes waveshaping. With a fast enough setting, we achieve hard clipping. IRC III does a much better job, but the modern and classic modes, when set to a fast setting, are essentially clippers.
Not a bad thing, but maybe not what you’d expect when using it.
FabFilter’s Pro L2 is similar - many settings are clippers unless you introduce lookahead.The lookahead gives the plugin additional time to determine the needed amplitude adjustment, which avoids the need for clipping.
The 4 functions that behave like a limiter - or a set ceiling without significant reshaping are the Aggressive, Modern, Bus, and Safe modes.
The Oxford Limiter does a great job unless we use a very fast attack and release time. But here, notice that we’re not clipping - waveshaping is occurring to lower amplitude aspects of the signal. Not exactly transparent, but at least in this case, we didn’t spend a lot of money on a limiter to get a sometimes limiter, sometimes clipper.
My personal favorite limiter, Voxengo’s Elephant, avoids waveshaping peaks in almost every instance.
If you select any algorithm other than clip, you’ll never achieve hard clipping. Even if you reduce the lookahead to its lowest value, you’ll achieve some waveshaping, but never hard clipping.
Furthermore, if you increase the lookahead to its highest value, even 20dB of attenuation won’t introduce harmonic distortion. Although this is also the case with FabFilter’s Pro L2.
Now, if all limiters did their job perfectly, they’d all sound the same. So a little variation is nice if you want your limiter to introduce a certain timbre.
But still, it was surprising info to find, and hopefully useful info for you all, that a lot of limiter settings result in hard clipping - and the ones I’ve shown here are some of the best digital limiters available. There are many more I’ve tested that are basically just clippers, regardless of the settings.
Watch the video to learn more >
I went into a deep dive about this in a video I’ll tag here - but in short, the algorithm used to measure loudness when determining how to normalize a signal, be it with YouTube, Spotify, Apple Music, and so on, can be tricked.
To do this, you cause the track to be measured at a lower LUFS than it's actually perceived, so that it’s turned down less - easier said than done, but here are the factors that cause that.
True-peak or inter-sample peaking is not measured during the test. So, any distortion caused by true-peaks, or added clipping during the encoding process from inaccurately quantized peaks, increases the loudness but is not measured as part of the loudness.
Next, 1-5kHz plays the biggest role in the measured loudness. If a track has a very dense 1-5kHz range, it will be measured as louder, and subsequently turned down more than a comparatively loud track with a sparse 1-5kHz range.
This explains a new but increasingly common technique in pop music, in which, for a large portion of the song, the vocal is the only instrument that significantly occupies that range.
Spotify, YouTube, etc will measure a track arranged that way as quieter. After normalization, it’ll be a full 1-2 loudness units louder than other tracks.
Now, personally, I don’t really care about this - unless a client really wants their masters aggressively loud, I don’t feel the need to introduce additional distortion for the sake of circumventing normalization.
But finding that there are so many blind spots in a very thought-out loudness measuring algorithm, probably even more than I was able to find, is strange and surprising.