Professional Mastering - Analog & Digital
Professional Mastering
251K+ subscribers
This is something that took me a while to figure out. Even after the various psychoacoustic classes I took in college, and researching for these videos, it took me an embarrassingly long time to come across this.
It’ll take me a minute to explain, but bear with me, it’s worth it.
It’s easiest to explain with an analogy. When we look at an EQ, we don’t think of it as having bandwidths.
However, it utilizes a finite array of frequency bands to group the signal.
With most modern digital EQs, these bandwidths are about 10Hz on average.
If I create a bell filter, and make the bandwidth as narrow as possible, and I move it to the low frequencies, you’ll notice I can center the filter on a specific frequency.
It’s difficult to see, but the filter is covering probably a couple of hertz in the lows.
As I move it to the mids, the distance between each frequency jumps to about 4Hz. At the same time, the bandwidth of the filter increases. This continues to increase until there are jumps of 10Hz, 15Hz, up to about 200Hz in the highs.
In short, this EQ, and most EQs, are designed to emulate how our ears perceive differences in pitch.
In the lows, our differentiation of pitch, about 1 semitone, is just about linear. But once we get to around 1kHz, increasingly drastic jumps in frequency are needed to hear pitch differences.
This is valuable to know in and of itself, but we can take this a couple of steps further and make it truly useful.
Just like an EQ, our ears break frequency ranges. As you saw, an EQ has thousands upon thousands of bandwidths - this is why I could center the bell filter on specific frequencies.
But our ears are not as good at this. Our ears can break up the frequency spectrum into about 24 bands, although research is ongoing.
These bands are called critical bands. If 2 or more frequencies fall within the same critical band, it’s incredibly difficult, if not impossible, to differentiate between them.
What often occurs is the perception of beating or modulation, sometimes described as roughness, since the same group of hair cells is being stimulated.
As you can imagine, the music and projects we make don’t care about our hearing limitations. Multiple frequencies often occupy the same critical band.
Furthermore, EQs, with their thousands of bandwidths, make it easy to affect the frequency spectrum in a way that doesn’t align with our psychoacoustic limitations.
Fortunately, some developers have done the research and picked up on this.
This Eqivocate plugin by Newfangled Audio does 2 things that fine-tune the filters to how we actually perceive sound.
First, you’ll notice that the Mel scale is used. Granted, this Mel scale or a similar logarithmic scale is used in other plugins, but this one actually centers the filters according to the scale.
This way, you know each filter makes changes based on our perception.
But here’s the really cool part - either intentionally or unintentionally, this plugin also mimics our critical bands through the use of very unique filter types.
In a 2019 paper, a strong relationship was discovered between critical bands and 1/3 octave filters.
If we look at this graph, we’ll notice how closely they correlate - the bandwidth of a 1/3 octave filter almost aligns perfectly with how we hear.
Meanwhile, Eqivocate uses 1/3 octave filters, with a unique pointed attribute - making it perfect for centering it on the critical bandwidth’s center frequency.
If we adjust the Mel frequencies, which already account for the logarithmic way we hear, and center them on the critical band center frequencies, we have an EQ that almost perfectly mimics how we hear.
Now, as I was adjusting all of these filters one at a time, I felt really clever - until I looked at the presets and realised that they’re already here, saved under the name Bark Bands.
So, let’s listen to these very specific filters. Notice how each time I solo a filter, it covers a range that sonically makes sense, so to speak.
Then, as I adjust the amplitude, that range becomes much easier to hear. No masking, no roughness - the unique pointed peak of the filter, along with its bandwidth, almost perfectly mimics human hearing.
Watch the video to learn more >
If the last chapter about critical bands felt like word salad, which it did for me even as I was writing the script, this tip is a lot more straightforward.
In short, whenever you get a new plugin, the biggest favor you can do for yourself is to try the presets.
I can’t tell you how much I’ve learned about audio, mixing, mastering, etc, just by going through the presets of a plugin and seeing how certain functions create a specific effect.
No one knows their plugin better than the developers, so using the presets they designed to highlight their plugin is almost akin to reading the manual.
For example, I’m using this new reverb plugin I got called Blurred, and at first I thought, well, this is pretty straightforward, I should just start making changes.
But, simple-looking plugins are deceptive and often have a lot going on behind the scenes.
It took running through the presets to understand how the algorithms interact with the distortion algorithms; how this sparkle function acts like an octave generator with some algorithms and not others.
Or how a fair amount of the reverse reverb function works well with the Emerald algorithm for a more modern vocal sound, but can behave entirely differently with a different algorithm.
So, even if you’ve been engineering for a while, presets are valuable. They’re not set it and forget it shortcuts for creating a sound, even if they do sound good from the start.
They’re instructions from the developers about how to use their plugin, from which you can learn, and then build on top of or tweak as needed.
Let’s take a listen to some presets and notice how, even with seemingly similar settings, the sound can vary drastically.
Watch the video to learn more >
Saturation has always been fascinating to me - you take a waveform, change its shape, and that generates new waveforms which can sound good if they’re harmonics or direct multiples of a note’s fundamental frequency.
On occasion, they don’t relate, and it doesn’t sound as good - although that depends on the project.
Anyway, the more you look into saturation, the more you realize it’s a very difficult thing to control.
The highest amplitude aspect will trigger the saturation. Even if you isolate the saturation to a frequency range, it’s difficult to ensure that the harmonics you want will form, since it’s difficult to be certain that the frequency you want to saturate will saturate.
This is where an emphasis/de-emphasis technique comes in. This means boosting a frequency range into a saturator and then attenuating by the same amount after the saturator.
Now, shameless plug, this is a plugin I’m currently working on. It’s a recreation of a popular texture circuit built on machine learning using the original hardware unit as the source.
It sounded good at first, similar to the unit I have, but it offered very little control.
So, I added this emphasis, de-emphasis filter. Something that you can do with any saturation plugin if you insert an EQ before and after it.
So, say I’m saturating a vocal bus, and I want a dense sound around 2-5kHz, which is often associated with vocal clarity.
To accomplish this, my first instinct would be to emphasize that range into the saturator and compensate after the saturator.
But remember, harmonics form above the saturated frequency. So, if you emphasize and saturate 2kHz, the next harmonic would be 4kHz if the saturator creates even order harmonics, and/or 6kHz if it generates odd.
So, the better approach is to emphasize half the desired frequency. If I want harmonics at 2,3,4,5kHz, I should emphasize 1kHz.
The 2nd order harmonic will be 2kHz, the 3rd order 3kHz, the 4th 4kHz, and the 5th order 5kHz.
Again, you can do this with 2 EQs and your saturator of choice, but let’s take a listen to the plugin I’m working on, since it includes this feature with automatically compensated emphasis/de-emphasis.
Watch the video to learn more >
Last up for this video, I wanted to share something I learned just recently.
I recently had a conversation with Fabien from Tokyo Dawn Labs, and he told me about something I had never even considered.
In short, when making changes in a DAW, each change takes a certain number of samples to register.
This creates a surprising amount of distortion.
For example, I have a 500Hz. sine wave here. After the sine wave, I’ll use this Prism analyzer and set the response to infinite so that I can see any changes that happen when I alter the gain of the sine wave.
As I change the level with the channel fader, notice how much distortion is formed around the sine wave.
Since the fader doesn’t make changes on a sample-by-sample basis, but instead every 4,5,6 or more samples, changing the volume mildly squares the waveform during the adjustment.
And this happens with just about any plugin or parameter.
As I adjust the pan of the sine wave, I’m also adjusting its amplitude to simulate the new stereo placement.
Again, the process of changing its amplitude occurs in steps due to the buffer size used in the DAW.
This doesn’t mean you shouldn’t automate settings - the distortion is pretty minimal.
But if you have maybe 20 or 30 tracks with automation occurring all at once, it can definitely alter the timbre of the track.
I can’t say if knowing this will drastically improve my mixes, but at least for future projects, I’ll be a bit more cognizant of how much automation I have occurring at one time.