How to actually master your own music
How to master your tracks to sound “right” instead of “enhanced”. Mastering your own music isn’t really a good idea, but neither is spending £100s you don’t have for someone else to do it. So here’s a guide on how to actually get it done, focusing on the why instead of of the what.
Remember this image:
This is how we’re going to think about mastering. Mastering isn’t about enhancement, it’s about getting the right shapes into the right holes.
Step one: buy really good headphones, but still spend 70% of your time using your monitor speakers
This post assumes you have a reasonably good pair of monitor speakers that you trust well, in a room with enough acoustic treatment to be fairly neutral from 400Hz upwards. It also assumes that your mix is good, and would translate well to other playback systems, before you start mastering.
Mastering cannot fix or even improve a bad mix.
But you don’t have a £2000000 mastering studio with £10000 speakers. How are you going to judge the low end and get a reliable second opinion on your sound?
It doesn’t have to be these ones, but you can’t go wrong with this particular model.
Step two: export your mix as a stereo track before mastering
There’s no technical reason to export your mix as a stereo WAV / AIFF file before mastering. You can master your song in the same session / project file that you recorded and mixed into.
But there is a massive psychological reason. When you’re not looking at an arrangement window filled with tracks, regions, channel strips, automation etc etc, your mix will sound completely different. I don’t know why. Maybe because your brain stops thinking “it looks like this, so it should sound like this” and instead pays attention to what it actually sounds like.
Export your mix as a 24-bit WAV or AIFF file, at the same sample rate you recorded it at, without any dithering.
When exporting, be careful that the tail end of any FX (such as reverb) doesn’t get cut off. This will not be visible in audio / MIDI regions in your track!
Some words on the “-3dBFS myth”
It doesn’t matter what the loudest dBFS peak of your mix before mastering is, as long as:
- The loudest peaks are less than 0dBFS
- It isn’t too quiet
Some people advise that your mix should peak less than some arbitrary amount below 0dBFS, such as -3dBFS or -6dBFS. The truth is, this doesn’t matter for exporting audio. It only matters when audio is played back.
I think the origin of this idea is inter-sample peaks. There is some fear that an audio file will “contain” them. But the truth is, inter-sample peaks only exist when your audio is turned into an analogue signal by your digital to analogue converter (ie, when the sound leaves your computer’s audio interface). And since we’re going to deal with intersample peaks later when mastering, it doesn’t matter if there are any (theoretical) inter-sample peaks exceeding 0dBFS in our mix audio file.
We will discuss inter-sample peaks in more detail later on.
Step three (optional): whatever distortion / analogue plugin is popular right now
This is an optional step, and is what some mastering guides spend ages describing. It’s not really that important, and depends on the character you’re looking for.
It isn’t a “secret ingredient” or “silver bullet” or “must have”.
There are many, many, many plugins that can add a nice bit of character. Everything from hard digital clipping (without antialiasing, which can sound surprisingly nice), to tubes, tapes, mixing consoles, vintage samplers, and pretty much anything else you can think of.
You only need to use one. Don’t stick 5 different distortion plugins in a row, each doing a tiny bit. Life is too short.
You have to use your ears and make your own judgement. Less is better than more.
I really like ToneBoosters Reelbus, with these specific settings:
ReelBus is the most practically useful of the tape emulators, as its interface is framed around what actually happens to the sound, instead of the buttons you would find on a real tape machine.
It adds a little bit of pleasing distortion, and also softens the high frequencies, but only on transients. I have “circuit clip” disabled, as I only want tape distortion and not distortion from the tape machine’s circuitry (which is a bit harsher). And I have “Spectrum” disabled (which is a just static EQ setting). The “Saturation” at 100% is what controls the amount of distortion and level-dependent high frequency compression found on tape (100% means the same as the tape machine it is modelling). I adjust the “REC LEVEL” to adjust how much saturation actually happens, and the output gain is automatically compensated for.
Step four (optional): “glue”
When the human ear hears a very loud sound (say, an explosion or a drum hit), the middle ear muscles involuntarily contract afterwards, for a very short moment. So things go quiet for a very short amount of time after loud noise.
Here’s where The Glue by Cytomic comes in. It’s a “bus compressor” but nobody seems to know why they really use it, other than that it “glues” your mix together.
Well, here’s how to actually use it: set the attack time to its slowest (30ms), and set its release time to either be fast or medium. Why? It makes drums sound more punchy, because it’s too slow to compress a drum hit, but immediately after the drum hit it briefly brings down the volume of the whole mix. Aside from drums, any sudden rise in volume of the overall mix will be met with a brief moment of quiet, emphasising it for dramatic effect.
It’s also true that it can make your mix sound more “full” just as any compressor does – but you don’t want to overdo this aspect of compression. It’s all about subtlety.
The downside of The Glue is this: It can narrow your stereo image if you compress too much. So don’t overdo it.
The Glue setup procedure
You can see my typical settings in the screenshot above. My process is this:
- Set the attack to its slowest (30ms)
- Set the ratio to its largest (10)
- Set the release time to its fastest (0.1s)
- Set the dry / wet mix to 100% wet. (We will be changing this later).
- Lower the threshold until you can really hear it “pumping”, but doing no more than 6-7 dB of compression
- Try out different release times to see what fits well with the music. I avoid using “A” (automatic) as it doesn’t pump enough, so I tend to stick with faster times.
- Adjust the sidechain high pass filter until the kick and the snare hits cause roughly the same amount of gain reduction, or until it otherwise sounds right (read this post to understand more about sidechain filtering).
- Adjust the threshold again until you’re getting a solid amount of pumping, but again no more than 7 dB of gain reduction
- Adjust the makeup gain, while wobbling the dry / wet mix knob to either extreme, until the dry and wet signals sound equally loud
- Adjust the dry / wet mix knob to taste (typically around 50%). The idea is that we get the glue to do a little bit too much compression, but then mix in some of the dry signal.
- Alternatively, use the glue in 100% wet mode, but with a gentler ratio and less gain reduction.
… And that’s basically it. Your mix gets a bit more punch and movement. It’s not a dramatic night and day difference, but it can sometimes work really well.
Step five: EQ (the actual mastering)
Nothing we’ve done so far is really “mastering”. It’s more what I would call “mix character improvement”.
But now we’re going to actually start mastering, and this is where it gets difficult. This is what many mastering guides don’t talk about, because it doesn’t involve buying a magical plugin, or using magical presets, or one-stop silver bullet techniques that solve all your mix problems in one fell swoop.
This is where you will really need to pay attention to what your speakers and headphones are telling you.
Find a reference track
First off, you need to find a reference track. This will be a critically acclaimed piece of recorded music, from a major label, that charted well, and that is vaguely in the same genre as what we’re mastering. (It really doesn’t need to be all that similar).
The purpose of our reference track is not to copy its tone or character, but to help us spot problems with our music.
The main problem when using EQ is that we get used to hearing our changes. So we need a “ground” to refer back to, in exactly the same way as you need a “ground” in electronics.
But we’re not trying to copy the sound of the reference track. In fact, our track may have a different tone, especially if it is in a different key, or has a different instrumentation.
Linear phase, analog phase, latency, minimum phase, IIR, FIR, frequency warping near the nyquist…?
Without going into too much detail about DSP, you can basically use Fabfilter Pro-Q 2 in any mode and it will work extremely well for mastering. You can also pretty much use any other EQ plugin. But there are some things to watch out for:
- If you do any parallel processing with an EQ in the chain, you should use a linear phase EQ, but sometimes can get interesting results with normal (minimal phase) EQ
- If you do any mid-side EQing, you technically also should use a linear phase EQ, but again, it may not be completely necessary.
- Some digital EQs suffer from frequency response “warping” or “cramping” near the high frequencies, which can make your EQing sound harsher. This isn’t a problem with Pro-Q (sort of). For more info, you can read Fabfilter’s help page on the subject, or read a blog post about it by vladg/sound: A classification of digital equalizers.
Warm up before you start EQing
Go to http://www.xhalr.com/ and spend a few moments focusing on your breathing. It’s not so much about breathing properly, but about becoming focused and aware.
Then shut your eyes or look away and listen to a few seconds of your reference track, before starting on your music.
Start with the obvious problems
There are some basic problems you might want to deal with first. Your mix may or may not have these issues:
Very low frequency rumble: Use a 12, 18, or 24 dB/oct highpass filter with a corner frequency somewhere between 20 and 50 Hz. But be careful here – if your music has intentional low sub bass in it, you don’t want to accidentally kill this.
Very high frequency rubbish: A similar lowpass filter with a corner frequency somewhere around 18 kHz. Why? many adults can’t hear much above 16 or 17 kHz. And the frequencies above this are mostly completely pointless. If you encode your music for streaming or as an MP3, the encoding algorithm can trip over itself trying to preserve this high frequency content that you don’t really need. Also, if you’re using any samples in your music, quite often your sampling plugin will have a very narrow spike of aliasing just before 20kHz, which will be inaudible to most, but we might as well lose it.
Stereo bass or low mids pulling the mix apart: Sometimes the mix will have too much stereo width in the bass or low mids. The easiest way to deal with this is to put Pro-Q into “Linear Phase” mode and change the Channel Mode to “Mid/Side”. Then insert a high pass (Pro-Q calls it “low cut”) filter, and change its stereo setting to “S” (side). A gentle high pass filter works best, and don’t go much above 250 Hz or so.
Don’t use steep filters. Steep high pass filters will ruin the tightness of your bass. And steep low pass filters can cause grainy-ness or ringing to your high frequencies. This is true regardless of what mode your EQ operates in, including Linear Phase mode.
After dealing with the obvious, sort out anything that’s “annoying”
This is probably the most important EQ step while mastering. Listen to your music (while referring back to your reference track) and try and work out if there are any frequency ranges that annoy you.
This could be very narrow resonant frequency spikes that somehow ended up in the mix, or narrow frequency ranges that just need a little dip, or slightly wider ranges that are just too prominent.
In general, the deeper the cut, the narrower the peak filter should be. Whereas shallow cuts can be anywhere between very wide or quite narrow.
But don’t go overboard. We want something subtle and gentle – it’s very easy to make your mix sound comb-filtered if you do too many EQ cuts. In general, if you end up cutting more than about 5 dB for a narrow filter, or 2 dB for a wide filter, you need to go back and fix your mix.
A common problem that I’m hearing everywhere
I’m hearing a lot of harshness in (non-commercial) mixes between 2kHz and 5kHz. People seem to end up with mixes where this area is too prominent. Probably because it makes things sound very bright and upfront. But people get used to it while mixing, and don’t realise that there is way too much. They end up with mixes where something sounds “off”, but can’t work out why.
As a rough guess, In 90% of cases where people want their mixes to sound both warm and bright, and think they need to reach for an analogue piece of hardware or plugin emulation, they could just make a small EQ cut somewhere between 2kHz and 5kHz and instantly get the warmth they are looking for.
Adjust the overall tone
Finally, use some wider peak / shelf filters to adjust the overall tone of your mix, referring back to your reference track. I’m talking about very gentle (0.2 to 2 dB) cuts or boosts. This step may not even be necessary at all.
The biggest obstacle to doing this well, is using filters that are too wide, too narrow, or in the wrong frequency range entirely. And that’s once you’ve gotten over the problem of “getting used” to an EQ change after you have made it. (That’s what we have the reference track for).
For example, many people will do a high shelf boost to bring up some high-frequency clarity and detail. But they’ll use a filter that also naively boosts some frequencies that should not be boosted. So you end up in a situation where the EQ always sounds wrong: because there’s not enough high end, or too much high end, but never the right amount. The same thing happens with bass. The key to this is to be very discrimatory in which frequency ranges you cut or boost – while at the same time not having any EQ which is too narrow or resonant.
Make sure that when you boost or cut, you’re not throwing out (or bringing in) the baby with the bath water. But also make sure your filters aren’t too narrow. It’s all about balance.
Finally: mid-side lift (optional)
A final technique to improve the sense of space in a mix is to apply a gentle, side-only high shelf boost. With Pro-Q in “Linear Phase” mode with Channel Mode set to “Mid/Side”, and the high shelf set to “S” (side), apply a very gentle (0.5-1.5dB) boost.
Final thoughts on EQ
It can take years of practice to learn to EQ effectively, and you never stop learning. But it’s one of the most undestimated skills in mixing and mastering. Because EQ is so difficult, people often immediately assume that it’s not their EQ skills that are lacking, but instead blame their approach, and then break out the multiband compressor, the “analogue” saturation / distortion effects, parallel processing, and all sorts of techniques that are cool but often completely unneccesary.
Step totally unnecessary: multiband anything
There’s nothing wrong with using dynamic EQ or multiband compression to fix problems that can’t be solved with “static” EQ. But you should try your best to fix these problems in the mix.
And that’s just it: multiband is very good for fixing problems. But the idea of using multiband during the mastering stage is something that seems to have been invented to sell plugins.
I’m not saying you should never use it. But the kinds of music you might think use multiband compression during mastering, do not.
And if you’re mastering your own music, then you also have the mix in front of you. And directly changing the mix is billions of times more powerful than using multiband during mastering. (You could even use multiband compression on individual tracks in your mix!)
And if your music gets played on the radio, the radio station will have their own multiband processing, defining their own “sonic imprint”.
The exception: other people’s bad mixes
Dynamic EQ and/or multiband compression can be extremely effective at fixing other people’s bad mixes. Use linear phase / dynamic phase mode (available in Fabfilter Pro-MB or Waves Linear Phase Multiband among others).
Step six: the actual limiting / clipping
For bringing up the volume of a mix, I like to use a very fast limiter / clipper that introduces some distortion. Why? Because when you limit your transients, you’re removing a lot of the punch & energy from your music. If your transients are distorted slightly, the distortion puts back some of the energy lost from limiting them. Typically, what you end up doing when clipping is slightly skewing the tone balance of your transients: where low to mid frequency peaks get clipped, their energy gets skewed towards higher frequency distortion which takes up less headroom, while preserving some (but not all) of the life.
I tend to find that clean, distortion-free limiters are problematic, as:
- When the release time of the limiter is short, but not short enough to distort, transients become soft and smudged
- When the release time of the limiter is long, you get lots of pumping, but aren’t really bringing up the perceived volume by very much.
So here’s my solution: Waves L1with its fastest possible release setting. It’s an old plugin, with a lot of dirt and grit. But in my mind, it’s the first “digital classic” compressor / limiter.
Don’t fool yourself with volume
In some limiters, the more limiting you do, the louder your master gets, and since louder things always sound better, it’s then hard to judge the damage that you’re doing to your transients.
Fortunately, in L1, there is a way of getting around this. Here’s the preset I have created:
- The threshold and out ceiling are set to -1 dBFS. This is so that there is some headroom in the signal coming out of the plugin.
- The release is set to 0.01 ms. In practice, it’s much slower than this, because L1 uses its lookahead to smooth out the gain reduction curve. But it’s fast enough to provide a small amount of pleasing distortion.
Notice that I’ve circled the “Link” button in red. If I click and hold this button, then drag up and down, it will adjust the threshold and out ceiling by the same amount. This means that when I lower the threshold, the volume won’t go up, and I won’t get fooled into thinking it sounds better. The perceived volume will either stay the same (when I’m just limiting transients), or become quieter (if I do way too much limiting). So I’ll be able to fairly judge how much damage is being done to the transients in the music.
What you want to do here, is adjust the Link level during the loudest part of your music, and see how much damage the limiter does to the punch of your music. (If the volume of your song before the limiter varies too much, then do some volume automation or slow compression before the limiter).
Step seven: the safety / intersample peak limiter
Earlier on I mentioned inter-sample peaks. What are they? If you really want to know the details, then you’ll want to read this wikipedia article, from which I’ve taken the following quote:
“The sampling theorem introduces the concept of a sample rate that is sufficient for perfect fidelity for the class of functions that are bandlimited to a given bandwidth”
Digital audio uses “samples” which measure the instantaneous volume level of sound at any given moment. Typically there will be 44100 of these samples per second. But what happens in between the samples? Using the power of maths, as long as you aren’t trying to record any sound which has frequencies over the “Nyquist limit” (ie, the sample rate divided by two), which can be enforced by simply putting a lowpass filter before your analogue-to-digital converter, then everything “in-between” the samples can be remembered and reconstructed perfectly.
In practice, however, it means that your standard digital peak meter, which shows the volume of the loudest sample in dBFS, may not show the real peak loudness of your music, because the peak level in-between the samples may be higher. These are what we call inter-sample peaks: and they exist when the sound comes out of your digital-to-analogue converter,
And in theory, intersample peaks are unbounded. This means they can be infinitely louder than the peak level of any of the digital samples. In reality, they usually aren’t that much louder, but loud enough to cause unexpected clipping in some situations.
That brings us on to ToneBoosters Barricade:
It’s my favourite mastering limiter, and is extremely good value for money. Here are the settings I use:
- The “AES17 + 3dB” setting is turned on, for correct RMS measurements
- ISP is turned on (intersample peak detection)
- Dithering is OFF. This is because my DAW can apply dithering when exporting the master, otherwise I would turn this on.
- Attack, release, lookahead, stereo link, and multiband are on their default settings, which work really well, but you can play around with these.
- I have my “Out ceiling” set to -1 dB. You can usually get away with -0.3dB or -0.1dB. I like to play it safe, as the loudness wars are over!
- I have the “Output level” meters on the right set to K12
- I then adjust the input gain until the outer meters (the thin ones, which represent RMS / average level) on the output meter hover around 0dB during the loudest part of the song. The inner meters (for peaks) will show much higher levels than this.
Step eight: volume automation
Finally, you’ll want to make sure that your song, when exported, starts and ends with a volume of minus infinity dBFS. You need to start and end with silence, to avoid any nasty clicks when the song starts or finishes. This may not be an issue, as in step two you may have exported a WAV or AIFF file that already starts and ends in silence.
But to make sure, it’s a good idea to insert a very small (1/10th of a second) fade in and fade out. Your DAW might be able to do fades directly on your song’s region. Or you might have to automate your fader. But make sure none of your faders in your mastering session go over 0dB! If your song starts with a drum hit right at the beginning, then maybe nudge your audio region forward by a 10th of a second or so, and make sure there are no accidental clicks at the beginning of the song.
Step nine: bounce 16bit dithered
Finally, you want to bounce your WAV or AIFF. Typically, you will want a 16 bit file with dithering. Not that I’ve ever been able to hear the difference between different types of dithering. 24-bit is a good idea for recording, but for a final master, 16 bit is more than good enough to deliver the final product. (In fact 11 bit would probably be enough, but 16 bit is the standard).
A 16 bit WAV/ AIFF is the lossless format that you can then use for any digital purpose whatsoever (converting to MP3 and other lossless formats, uploading to streaming services, etc).
Step ten: listen on lots of devices and check you are happy
This is the final step, and, unfortunately, you will need to get over the fact that your music will sound different on different systems. But it should always sound good and “clear”. And remember, people who listen on terrible speakers are used to what those terrible speakers sound like, so if you listen to your song on your friend’s £30 2.1 speakers and it sounds “unbalanced”, don’t panic, and remember that that’s what all music sounds like for them.