It's no news that most music listening nowadays happens via streaming sites, and applications like Spotify, Apple Music etc, as opposite than by picking up a CD and putting into a player.
And no surprise: it's eminently practical! When my five years old son asks for the 100th playback of Eiffel 65's I'm blue (don't ask), it's a breeze to take the phone, tap a couple places to stream it to the wireless speaker, and off he goes, dancing to his heart's content (on second thought, maybe I should actually revert to CDs to make it harder to play that song.. but I digress).
Point is, these days I use my CD player almost always for referencing, except he occasional full blast of Vivaldi's The Four Seasons when nobody's around. To each their own.
Most "regular" listening happens via streaming. And streaming would be perfect if only they paid artists decently... but that's another post: streaming is here to stay.
That given, the question is: does this make any difference to the type of masters we want?
Say we have mixed a track, and we sent it to a mastering house, and they ask if we want to pay a little more for having it optimized for streaming. Is it worth it?
Let's start from the beginning: a common thing about all internet-based music services is that, when you upload your CD-quality 16 bit, 44.1 KHz master, they go and apply lossy compression to it, in order to reduce its size. And that is the file that will be streamed.
All lossy encoders work pretty much the same way. They are rooted in Fraunhofer's work: their researchers spent almost four decades finding out how to cut away information from a PCM stream to reduce its size without damaging the perception of audio too much, culminating in the 1990s in the Mpeg-Layer 3 lossy encoding scheme, with its psychoacoustic model baked into it.
So, short of little differences (and maybe different ways to expose tweaking parameters to the user), all good encoders do pretty much the same job. One of the best mp3 encoders, incidentally, is a Google search away: it's called "Lame", it is open source, and can be used by anyone.
The quality of the encoded (i.e. lossy-compressed) audio always depends on two aspects:
the target bitrate for the encoded file (the more,the better, but the larger the file);
and the amount of artifacts that may occur when reconstructing the original waveform. Such artifacts depends on the fact that - unlike with "normal" sampling - when audio is reduced in size in a lossy way, we actually lose information; so the reconstructed waveform may differ from the original uncompressed one. These artifacts depend on both the process but also the actual content - so how much quality is lost (or kept) depends a little on the audio itself.
Is there something we can do at mastering time to optimize our final track to preserve quality even when (inevitably) will be lossy-compressed? And, since we are at it, do we need to pay a little more for a supposedly special mastering process (Apple for example has "mastered for iTunes"?
The short answer is yes: ensure your master peaks at most at -1dBFS (and not 0)
Let's assume we have a good encoder (such as Lame, or Fraunhofer's, or the AAC one built by Apple).
From the above, it's obvious that in order to preserve quality, encoding should occur with a high enough bitrate. Unfortunately we do not control this, since the bitrate is decided by the streaming platform.
Therefore, no "special mastering" can help with the bitrate. Either a streaming service encodes with higher bitrates, or it doesn't: if your track is sent to a "SuperFastify" streaming service which consistently encodes at 32Kbps, your track won't likely sound that great no matter what kind of master you upload. Sure, Apple's AAC encoding may keep a better quality given the same file size reduction, but that doesn't help when we upload to a non-Apple platform which may not use it.
That leaves us with what we can control: the properties of the master in terms of dynamic range, max peak level and overall equalization.
Still to this day, we tend master for "CD-quality" - which nominally means PCM format at 44.1KHz sample rate and 16 bit word length. But "CD quality" actually means something more: masters intended for printing to physical CD and for reproducing on an actual compact disc player, have no problems peaking at 0 dBFS.
Any good quality CD player is specced to handle perfectly well the full dynamic range possible with 16 bits, in perfect fidelity (no distortion, no artifacts), so it's perfectly ok that the master peak level touches zero decibels full-scale. And due to peak normalization and loudness wars, very often CD masters are quite heavily limited and tend to reach that peak quite often, and generally stay in the top part of the scale.
But nowadays, very few tracks will ever see a CD.
And even if they do, they will also see a lot of streaming platforms, and many of these platforms will use loudness-normalized playback.
In this case (which is basically every case nowadays), to keep a good quality a PCM audio master must peak quite lower than 0 dBFS. Easily go down one decibel, and peak no more than -1 dbFS.
Why? Because the audio waveform will be reconstructed from less information than what was available in the original PCM file. Since we have less information, it may well be that, in a place where the original PCM waveform peaked at exactly 0 dBFS, the one extracted from the encoded file ends up over that (what is called an intersample peak), causing digital clipping - which sounds really bad. That really depends on the material.
In practice this means that a PCM master good for a streaming platform will usually be a little less loud than if it was intended only for CD playback.
However, since most streaming (and many broadcast) services are loudness-normalized (or will be), this is not a big deal. Peak level matters less in such environments.
Is there something more we can do when it comes to the audio itself - to make sure it comes relatively unscathed even if the streaming platform uses low-ish bitrates?
Well, in very generic terms, one of the things that happens with lossy compression is that the encoder will try to "save" the frequencies we hear the most - i.e. the midrange. If something needs to be cut, it'll be lows and highs first.
That means that to survive low bitrates (to the degree that it's possible) the master's balance should not be too heavy on these frequencies.
That does not mean that you shouldn't have thundering bass and shimmering highs (if they're part of your track, you obviously should!), but to use mixing practices to move the "feel" of them to the midrange rather than rely only on equalization.
It's very much the same idea with which way you preserve the sound of a mix when it's played back on heavily band passed systems (laptops speakers, phone speakers, cheap earbuds and the likes): you use psychoacoustic trickery at mixdown time to "move" information to the midrange, tricking the listener's head to "hear" frequencies that, physically, aren't there. Tricks of the trade are stuff like exciters (to add bass and kick harmonic), selective muting (to isolate high frequency sounds and make the stand out), keeping a wider dynamic range (to make differences "pop" more) and so on.
Some of this can surely be applied at mastering time. But an arrangement and mix which keep their feel as much as possible when played on less than stellar systems, will likely survive streaming better than mixes which sound great only on very good playback systems.
It wasn't that different with vinyl, that for physical reasons could seldom reproduce all the bass found in the master tape (otherwise the stylus would jump!) and where the high frequency detail was limited by noise, enormously more than with digital recordings (encoded or not). So you had to "move" it a little in the mid range, using the same trickery. If the mixing engineer kept that into consideration
So, in conclusion, do you need to make special masters for streaming?
If you're mastering for loudness rather than peak level, it's very unlikely. A master that "works" on a iphone, car, home higi, PA and disco-system (which is what a master is supposed to realize) will work fine for streaming.
If you're really aiming at printing and selling CDs, you may want a separate CD-master which can be a little louder (peaking at 0dBFS instead of -1) and can keep more "real" low end and highs and be a little more limited (aka less dynamic).
Most of the music nowadays, especially in home studios, will be squarely aimed at the streaming market, so generally you won't.