The Metadata Blame Game

Roon Software Discussion
24

I have been a user of open source and free software for many years. The IT world as we know it wouldn’t be the same without the often voluntary and unpaid work of many private developers. I therefore would be more than happy to be able to offer some of my free time to input structured metadata of performers, productions and musical works. For me this would be part of the musical hobby and, of course, I would concentrate on artists and albums I really care for. Even if this work contributes later to business opportunities for third parties. Around the universe of open source software there also has sprung up an industry which takes advantage.

At the same time, I must say that I also understand the feelings of @Jazzfan_NJ that Roon just isn’t doing enough. I often feel the same; to me it seems that there has been a lack of creativity and proposals to find some solution to the thorny metadata issues which have already been touched upon in this thread. Roon’s marketing leans heavily on its superiority with regard to metadata presentation and linking. Of course, that there are musical genres where complete and correct metadata is sorely missing, goes unmentioned. And yes, is has seemed far too easy to simply put the blame on the next actor in the supply chain, without ever coming up with some good idea for how to go forward…

So, here’s my seed of a proposal to advance, instead of keeping still in the face of a seemingly herculean task.

I’d love to see Roon offer on its artist, album, composer and composition pages links to a web form, preloaded with the available (possibly incomplete and/or wrong) metadata. This form should allow Roon users to edit and input metadata in a structured way, conforming to a standard, for which I can think of MusicBrainz. Rather than tell users ‘go to MusicBrainz if you miss metadata’, offer us some easy-to-use and friendly interface, without forcing us to engage directly with the finicky MusicBrainz editing and reviewing standards and editors.

The second step of the process would revolve around Roon staff reviewing and submitting the metadata editions into the MusicBrainz database, for benefit of all. There would be one point of submission into the MusicBrainz database, managed by experienced and paid Roon staffers.

I’d also love to see Roon come up with some idea for making their users feel appreciation for the work they do. I remember all too well the Valence picture craze, which nowadays seems all but forgotten and a thing of the past. I remember the incredulity as in the first days and weeks many more pictures were submitted than anticipated, even by the most optimistic staffers. But this simply ebbed away, and I don’t remember having seen some thread for continuous feedback and appreciation. There could be some monthly ranking of most edits submitted, and some offer of appreciation for the first three in this ranking… something like that…

As I said, this is just a simple thought, a seed, and you can water and fertilize it and, possibly, try to make it grow big and bear fruit, or simply step on it and make it go away…

17 Likes
25

Perhaps this is a view of the situation through the lens of frustration. Think of it this way: Lots of non-profits sell things people have contributed, in order to continue their work. And it comes back to you, through your Roon subscription (which is, after all, a sunk cost).

I painstakingly detailed many jazz releases to the CDDB database, and was outraged when GraceNote took things over. But even then, many copies of my metadata contributions made it out there.

1 Like
26

I did a great deal of work some years back on parsing of raw metadata found “in the wild”. It’s not a simple thing.

And the key fact about “AI” is that there’s no “I” there – it’s all “A”.

5 Likes
27

I really like this idea! I agree with you that adding to MusicBrainz directly is finicky at best, and the approach you propose scales much better. I’d add that the initial preloaded data could be complemented with metadata from the ID3 tags of the material the user is trying to annotate (which is what the MusicBrainz Picard app does in its clunky way). This could all work as a “Suggest metadata” screen when “Identify” cannot find a match for a newly uploaded album. With such a feature, I’d do this a lot more often than with Picard, which requires me to go through a lot more steps.

6 Likes
28

I also think that there must be a MusicBrainz API which would allow to make the final submission of reviewed and signed-off metadata edits a daily automated batch job.

This could easily be a project where both MusicBrainz and Roon have something to win… leverage the combined musical expertise of hundreds of motivated Roon users by an easy-to-use, intelligent and friendly interface, and see the MusicBrainz database grow and getting debugged like never before… :smiley:

3 Likes
29

The way Picard works is to prefill MusicBrainz submission pages, but then quite a bit of editing may be needed. For example, an artist in the album may not be known by MusicBrainz, so the artist must be added separately. Various edits may also be needed when multiple artists are involved, to make sure they are handled correctly in different fields. And so on… I’m skeptical that purely automatic API entries would be allowed by MusicBrainz, as a lot seems to depend on human judgment to collect and organize missing or poorly formatted information.

1 Like
30

This seems like the ideal solution to me too, lower the friction and leverage the interface Roon users are already using in order to fix metadata and submit additions back to MusicBrainz without a user having to think about it.

If indirectly MusicBrainz (a non profit) leverage my changes and addtions, which I’d have made for my own use anyway, to keep the lights on, so much the better.

Roon could also leverage its large user base to help automatically validate (or at least appraise) the data before submission

For example if multiple users have suggested the same metadata it can probably be submitted to MusicBrainz with a high degree of confidence.

Likewise, Roon could also (optionally) display the recent additions and edits from their user base (with an indication that the metadata is so far unvalidated, but as a user you are invited to approve or disapprove it). Once X users have approved or disapproved you have some metrics to help decide whether to submit it back to MusicBrainz.

Additions from users who regularly submit data that aligns with others or whose submissions are regularly approved by peers may need less validation in future before one of their additions or edits is submitted back to MusicBrainz.

Harman (as a wider company) could possibly even right off some tax by working with MusicBrainz (a non profit) to support an initiative to pay for some humans in the loop to spot check and validate the incoming data.

But for me and I suspect others, the pill sweetener here would be that your edits are being fed back to a central, non-profit like MusicBrainz for wider use outside Roon.

.

3 Likes
31

I use Picard all the time to add new stuff to MusicBrainz. This is for Plex as well as Roon or anything else that taps into it. The arbiter of MusicBrainz though can be a tad touchy on certain things but then let others glaringly wrong slide.

2 Likes
32

Indeed, despite my post above, i know from first hand experience how very quickly it can get complex.

Your example of first having to add an artist before you can link to them is a good one, but it barely scratches the surface of the complexities here. Is that artist already in the database? How can you validate that?

You can use weightings to gauge the likelihood of metadata being correct, but at the end of the day you need expect users to be sure, Even then you’ll have times when expert users will disagree amongst themselves as to what is correct.

On a more positive note you have to start somewhere and maybe just a link to the relevant MusicBrainz submission page from Roon is an extremely low hanging fruit that might encourage some users to make edits who previously weren’t aware they could.

There is also a lot to be said for, and gained by, MusicBrainz working with Roon (and others) to build better forms and interfaces (hosted at MusicBrainz) that help on-board new users without overwhelming them.

I know from similar projects I have been involved with, that experienced users often forgot how hard it is for new users to get started, how confusing and scary expert interfaces can be to the uninitiated. Complex forms and processes can obviously help stem the flow of low quality submissions, which the scarce expert users then need to clean up anyway, but it also risks scaring off users who might be the expert users of the future.

In short, I wholeheartedly agree, there’s no Magic Bullet….but there probably area’s where small incremental improvement for be made and friction reduced.

3 Likes
33

This is actually an interesting topic overall - the short answer is that genre metadata is more about marketing than about the actual reality of the music. I’m going to risk sounding like I sniff my own farts a bit. A genre is very loosely a description of the arrangement of a piece of work - but in the modern music environment with modern editing, modulation, etc arrangement is not limiting.

Importantly, genre is a severely lagging indicator. We aren’t in the 90s anymore where you hear rumblings in the distance of something someone calls ‘grunge’ and then suddenly an army of rockers wearing flannel appear driving all before them. As early as the 2000s we see that throughout Europe hundreds of relatively well defined ‘sub-genres’ are arising across Metal and Electronic music defying any kind heirarchal analysis - Drone Metal and Blackwave end up being the same ‘genre’ convergently evolving from two parents.

Obviously a lot of these are really niche genres - but well defined small genres are just one side of the coin. The other side of the coin are hit ‘genreless’ artists. Where does Mumford & Sons live? The metadata says Alt Rock/Indie Folk but the average Indie Folk is going to be something like Bon Iver and the average Alt Rock is going to be Vampire Weekend. Bon Iver + Vampire Weekend is not equal to Mumford & Sons.

Mumford and Sons is arguably a relatively novel strain of Pop-Folk borrowing heavily from Punk where the Punk ends up being a stylistic flair of the mainstreamy Pop-Folk as opposed to Dropkick Murphys who are primarily Punk with folk styling. In fact I’d say we can see this stylistic evolution where we see that capital-P punk has kind of died off and instead we see that -Punk is kind of emerging as the new world where ‘punky’ arrangements are ending up as a subtype of a variety of genres. Emo-Punk becomes Midwest Emo/Pop Punk, Country Punk is surging with people like Oliver Anthony, Folk Punk with Mumford and Sons.

This is a lot of talk but the takeaway is that any given album fits one of three classes:

  • Specifically Genred: This album is intentionally made in the style of a genre such that if the genre data is missing it means one or more professionals along the way have just not done their job to mark the genre
  • Novel Genre Fusion: This album is intentionally made to fuse two genres together in a way that is currently not common (Punk-Opera anyone?)
  • Non-specifically Genred: This album’s artistic vision doesn’t intentionally represent a specific genre.

Only the first case can be resolved via AI - but we would also need to be identify a given album’s membership in the group of albums that are genred but people have failed to add the editorial metadata.

All of this and the easiest way to make someone mad is to tell someone that a given album fits in a genre when they disagree. Nickelback is metal.

5 Likes
34

Now I feel lost… the discussion here was about the lack or insufficiency of metadata available for new releases and, for some less popular musical genres, also for old releases. It’s about album and track metadata, artists, production personnel, production data, etc etc…

Was all this about fitting releases in one or more of musical genres??

2 Likes
35

I thought the same and am more or less entirely in agreement with @nathan’s genre exegesis, but as @Andreas_Philipp1 says, the topic is much broader and @Jazzfan_NJ’s example contained mostly instrumentation credits and a release date, and one may wonder if AI can’t extract it

36

not true. there are standards, and paid editors at the labels and the various metadata providers (excepting those that are crowdsourced.)

37

To support @WiWavelength here, even for well-known labels in certain genres, metadata is woefully incomplete.

1 Like
38

Thanks for this. I like weird genre combinations and, sometimes, I do find some combinations most entertaining. From '21… still looking for streaming genre search :slight_smile:

39

As Sr Director of Sales for a big commercial metadata provider, I am sure @woodford knows about the lack of metadata, especially for some less popular musical genres. He seemed to wish to point out that those commercial metadata providers pay for human expert editors, and crowd sourced providers like MusicBrainz don’t.

I can’t know to which extent a big commercial metadata provider like Nielsen may overcome this problem within Roon, and how this may impact Roon’s cost structure and prices for end users, and if this would make user contributions unnecessary. I want to believe that Roon may have considered this option, though.

40

This would work if everybody was to agree on the exact content of each of those genre definitions. Problem is, the narrower the definition (subsubgernre), the less agreement on where a track/album should be put.

So you might enter a combination of genres according to your view and miss what you are looking for, because others look at it differently.

This is mitigated by fewer and broader genres, which are not up to your task.

2 Likes
41

The problem we are talking about is quantity – many releases have zero or little metadata – not quality. Expert editors don’t do much for quantity as far as I’ve been able to observe over decades of paying attention to this.

42

I’d argue we are looking for both quantity and quality… I’ve no idea whatsoever what commercial metadata providers have on offer, but I think that if Roon implemented an interface with MusicBrainz for crowd sourcing metadata edits, both quantity and quality could be addressed. I’d happily be part of this effort…

1 Like
43

Nickelback is metal haha loved this.

They certainly don’t belong with roadrunners records that’s for sure.