There is a seismic shift underway in the manner by which consumers interact with their music. We can feel the first tremors, and some of us are even starting to get our first glimpses of what the post-shift world might look like. But for the most part, like with all paradigm shifts, we are only ever going to be able to fully comprehend it with the benefit of hindsight. You think I’m talking about things like TIDAL and Spotify, but I’m not. I’m talking about data.
All of our music has data associated with it. Whether we own the disc, LP, digital download, or even if we are just streaming, there is a lot of data associated with that music. Historically, most of that data has been provided on the LP sleeve or CD booklet, and includes information such as who is playing on the recording, where it was recorded, who the producer or balance engineer was, who wrote the music, and so on. Other data exists elsewhere (for example, on the internet) and might include such nuggets as, which other recordings the producer produced, when and where the performer was born, who else has recorded the same music – and even what a balance engineer is, and what he or she actually balances.
In the brave new world of digital audio, the technology already exists – it has long existed in fact – to put all of this data in a place where its relationship to the music can be cataloged and accessed on demand. It is what we term metadata, and includes Album, Artist, Track Title, Composer, and so on. This information is embedded into most digital audio file formats where the playback software we use can read it and make use of it. In the screenshot below (taken from iTunes) every single thing you see – every word, every picture – has been extracted from the album’s metadata.
This metadata is what actually makes the process of using iTunes as a tool for music playback at all possible. Of course, we don’t see Tilson Thomas’s birthday listed, nor do we see who the balance engineer was. And we’re still none the wiser as to what the balance engineer actually balanced, or if anything was even balanced in the first place. But it’s a good start.
On the other hand, there are some things that, intuitively, you might like to do, but can’t. For example, I would like to be able to click on “Michael Tilson Thomas” and be shown a list of other recordings on which this person appears. I would like to be able to click on “Symphony No 7 By Mahler, Gustav” and be shown a list of other recordings of this work. I would like to click on the album art and be shown a PDF of the CD booklet so I could browse through it. I would like to be advised that “Michael Tilson Thomas” is the conductor, and that “San Francisco Symphony” is the orchestra, instead of having to rely on my own common sense to draw those inferences. All these things are not only feasible, they actually present no technical challenges whatsoever if Apple wanted implement them. The seismic shift will occur when those things – and many others like them – actually start to happen on a routine basis.
At this point it is fair to ask what is preventing any of this from happening right now. Part of it is inertia – that’s the bit that stops Apple from writing the code and pushing it out with the next major update of iTunes. But that’s being unfair to Apple, because even if they did write that new code and roll it out, it would only serve to emphasize the main obstacle is really the central issue to all this; the one that is holding back the paradigm shift from actually happening … access to data. You see, there is no straightforward way for iTunes to grab that data – especially if it’s not in your files’ metadata. And very little of it is.
Right now, the metadata that is found in digital audio files is typically limited to the track title, the album title, the track and album artists, the composers/songwriters, the year, the track number, and some user comments. This is because these formats have arisen in an ad-hoc manner, primarily reflecting the personal wants of the original developers, rather than through serious consideration of the implications of its adoption in a comprehensive and global manner. Concerted efforts have been made from time to time to expand this limited metadata into what I term “rich metadata”, but these efforts have always run into their own dramatic, and generally show-stopping drawbacks. However, that is not for this column.
This column is about where all this rich metadata comes from in the first place. Because, the 800 lb gorilla in the room is that it doesn’t come from anywhere. While the music industry has indeed published all this information – historically in the booklets that come with CDs – it has not typically made it available in any indexed or searchable form. This is important, because if I have the capability in my digital audio files’ metadata to store the name of the track’s balance engineer, then there needs to be a way I can access an online source and specifically request this item of information. So if one source decides to publish it as “balance engineer”, and another publishes it as “engineer (balance)”, and yet another as “Recorded by”, and so forth, then the process of extracting that information is going to be awfully cumbersome. Indeed, another source may publish the information as “Other”, and it is not until you retrieve that data set and find an entry which reads “Fred Blowthrough (balance engineer)” that you have an opportunity to extract this information. This means you can’t just retrieve the “balance engineer” entry, but must instead retrieve all of the “Other” entries and post-process them afterwards.
So for rich metadata to work, a realistic prerequisite is for the data to be cataloged in a comprehensive, systematic, and unambiguous format. Right now, there is no such format that everyone has decided to agree upon. Which is not the same as saying suitable database formats don’t exist – they most certainly do – but there isn’t one that the industry has broadly agreed upon to be the standard. So you might well ask, if metadata is so important, then how come the situation is such a mess? Well, it’s a combination of things.
There is the fact that putting together such a database represents a phenomenal amount of work, and in return for doing all this work most people tend to have the reasonable expectation that they should be able to get paid for it. Gracenote and Rovi are examples of rich metadata databases that are accessible on a fee-for-service basis. Roon is an example of a business that has essentially wrapped a product interface around a rich metadata database as a means of monetizing it. There are others. But these vendors offer their service at a premium price, one that anybody who downloads music player software free of charge is not going to consider paying for one minute. And none of those can currently be integrated into your software player of choice – take iTunes as a simple example – even if you were willing to buy a license to do so. Finally, even those premium services – with their premium access fees – do not necessarily deliver total data coverage. In reality, they are not significantly better (where significance is evaluated with due regard to the $$$$ you have to pay to access them) than the free service MusicBrainz, which I am going to discuss in some detail.
Where do those premium data services get their own data from? Many – maybe most – of the serious record labels out there have internal databases which store a lot of this information. But each label’s database is private, in its own format, and stores only the data that it has decided it wants for its own internal purposes. It is an inward-looking capability, designed solely to meet its own needs, not an outward-looking one designed to meet someone else’s needs. In addition, let’s not forget that this is the music industry we’re talking about here – not exactly the most enlightened business leaders on the planet. These are the same people who still think in terms of their digital studio masters being “Crown Jewels”. So it is not at all clear that they actually have any interest whatsoever in cooperating with the process of populating third-party music metadata databases. If you are a premium music metadata provider, you therefore need to be scouring the catalogs of every single record label or music source out there, and reformatting the data you find into a consistent form within your own database. That is a huge challenge, and one for which there are no ready-baked solutions that I am aware of.
However, all is not lost. There is a major, free-access music metadata database out there, called MusicBrainz. It crowd-sources all of its data, and has been around since the turn of the millennium. Its origins lie in what many still see as the CDDB/Gracenote fiasco. CDDB (which changed its name to Gracenote) was the original crowd-sourced music metadata pioneer. However in about 1998, it took the controversial decision to start charging a fee to access its data. Since this data was all crowd-sourced, and unpaid, many people felt aggrieved by what they saw as little more than brazen corporate appropriation of public property. Today, Gracenote is a very successful enterprise, and has expanded the scope of its database activities into video, sports, and even automotive areas, and is a part of the Nielsen group (of TV ratings fame).
MusicBrainz was essentially a spin-off of the crowd-sourced aspect of CDDB/Gracenote, and continues to operate in that format today – crowd-sourced and free-access. When I first started looking at MusicBrainz I was not particularly enamored of what I saw. Being a crowd-sourced activity with precious little in the way of a budget, their promotional efforts are not as slick as you might be used to seeing. It comes across as clunky, difficult to use, and the community has a kind of hacker vibe, dwelling mostly as it does on IRC. If you want to learn about MusicBrainz you really need to dig into it seriously. I started to do that in January of 2017, and as a consequence I have now developed a huge respect and admiration for what MusicBrainz has accomplished. The people at MusicBrainz have thought through just about every aspect of what a music metadata database should be, and the result is surprisingly close to flawless. Where there are apparent flaws, these are mainly the result of the crowd-sourced nature of the project, and the fact that crowd-sourced data must always end up reflecting any flaws which have crept into the norms of broader usage.
Over the next couple of editions of Copper, I will be digging into MusicBrainz in some detail, and looking at how a resource such as this can both enable the development of a high-performance music server product, and at the same time limit its effectiveness.