Horrible edge cases when dealing with music

https://dustri.org/b/horrible-edge-cases-to-consider-when-dealing-with-music.html

by pabs3 • 4 years ago

533 189 4 years ago

f7ebc20c97• 4 years ago

I'm gonna name my band an 18 TB string of random characters, interspersed with an "idea" that you can't even put into text! Take THAT, Unicode!

javajosh• 4 years ago

Ooh, that gives me the idea to name my album the base64 encoding of it's tracks. The album itself, of course, would be empty.

akira25012• 4 years ago

I believe you would no longer be able to copyright the track, then. So, you should encrypt the encoded titles, but record an audio track that is a voice recording of you reading out the encryption key for that track. You know, for DRM.

eyelidlessness2• 4 years ago

Hi are you Fiona Apple and if so can we be friends?

scns3• 4 years ago

Is that the limit on Postgres' TEXT column?

petsounds• 4 years ago

There is a band from Pasadena, California called Ozma. Way back in the day (1999), they released an album called “Songs of Audible Trucks and Cars” [0] via MP3.com. I later found out that they really wanted the title to be “Songs of Inaudible Trucks and Cars” [1], but MP3.com had a character limit on album titles, so they had to shorten it

[0] https://www.discogs.com/release/8369169-Ozma-Songs-Of-Audibl...

[1] https://www.discogs.com/release/15809071-Ozma-Songs-Of-Inaud...

revolvingocelot• 4 years ago

If, like me, you immediately thought "why not just drop the 'and' and replace it with an ampersand?", Ozma had to do that, too, and parent's quoted the album title incorrectly, alas.

tashi• 4 years ago

"Songs of Inaudible Automobiles" would have fit and been fun to say aloud.

throwaway684936• 4 years ago

Presenting my new album: SQL Injection'); DROP TABLE Artists;--

Tokkemon2• 4 years ago

Amazing!

aaaaaaaaata• 4 years ago

Provocative, you could say.

porcoda• 4 years ago

In addition to all of these edge cases, it doesn’t help that the metadata services have a lot of inconsistency as well. Say I have a song X that was performed live on DATE at some VENUE. It may show up as titled “X”, “X [live]”, “X (Live DATE)”, “X (DATE VENUE)”, and so on. Stuff that should be in metadata ends up in strings embedded in song and album titles, which makes for a big mess. I curate a huge collection and trying to keep my metadata consistent is a massive headache. I think I’ve tried just about every tool out there and still haven’t found one that I’m happy with.

logbiscuitswave• 4 years ago

Or the many ways to have “featuring” — “feat.”, “w/“, “ft.”, “(feat.)”, “[feat.]”, etc. MusicBrainz does a laudable job trying to standardize all this crap and as someone who is really particular about how their music metadata looks it’s highly appreciated.

quickthrower2• 4 years ago

No mention of Prince?

nocman• 4 years ago

You mean "The artist formerly known as the artist formerly known as Prince"?

:-D

joshspankit3• 4 years ago

Another great one for the history books. His symbol broke so many music company systems and the story earned the respect of many.

seaish• 4 years ago

Related, songs that are meant to be played by accident: https://gimletmedia.com/shows/reply-all/j4he7lv/183-the-veno...

Example album: https://drumkoon.bandcamp.com/album/they-tried-to-ban-this

He has hits like "Hey Gugle Play Music" and "Hey Siri Play Space Music".

mherdeg• 4 years ago

I enjoyed his track "Sorry Katie I Spammed Your Smart Speaker".

bobbyi• 4 years ago

Amongst major artists, Tupac is good at causing confusion. Is he Tupac, 2Pac or Tupac Shakur? Does it depend on the album? Is his alter ego Makaveli a different artist?

emptybits• 4 years ago

Yup. ARTIST_ID does not have a 1:1 relationship with ARTIST_NAME over time.

Another example... from 1976 to present day: Johnny Cougar → John Cougar → John Cougar Mellencamp → John Mellencamp.

Close enough for library software to store/recall them all as "John Mellencamp"? Maybe. But perhaps not to fans or historian or artist liking.

geenew• 4 years ago

Richard D James == "Aphex Twin" || "AFX" || "Caustic Window" || "Bradley Strider" || "GAK" || "user48736353001" || "Polygon Window" || "Universal Indicator" || "The Tuss" || "Powerpill" || "Mike & Rich"

(for me, 'artist' is the alias, 'album artist' is 'Richard D James'...)

webmaven• 4 years ago

Don't forget Garth Brooks|Chris Gaines and Eminem|Slim Shady

Klaster_1• 4 years ago

Imagine documenting something like this https://www.discogs.com/artist/661099-Filipe-Santos (check out the edit history) where the artist actively sabotages the documentation effort and makes their music unavailable on a whim.

tzs• 4 years ago

It's not much an edge case in the sense of those, which mostly concern software authors making incorrect assumptions about the data in fields, but rather a "not even realizing we need a field" case, but music library managers are often quite annoying when dealing with classical music.

They really want to fit everything into the artist/album/song scheme, and that doesn't really work well for much classical music. In classical music most of us would want something like composer/piece/orchestra/conductor/performers (the "performers" field would be for performers in addition to the normal orchestra, such as soloists or choruses), and it should be possible to change the sort key order among those fields.

jay-anderson• 4 years ago

Definitely. Pieces end up being listed something like "Symphony No.3, Mvt. II", but they only show the performer which is the orchestra playing rather than the composer so I have no idea which symphony it is. I haven't seen any service get classical music right.

I think your list is pretty good (composer/piece/orchestra/conductor/performers). It'd be nice to also have a field for the movement. Also I totally agree with custom sort order being a must. E.g. for choir pieces swap out 'orchestra' for 'choir', for piano pieces only include 'performer' field, concertos should have the 'soloist' before the orchestra, etc.

CharlesW• 4 years ago

…music library managers are often quite annoying when dealing with classical music.

Apple will apparently be solving this problem this year with a dedicated "Apple Classical" app, the origin of which is their Primephonic (https://www.primephonic.com/) acquihire.

"Why are you shutting down Primephonic?"

"To focus on creating an even better experience for customers around the world. Apple Music plans to launch a dedicated classical music app next year combining Primephonic’s classical user interface that fans have grown to love with more added features."

nerdponx• 4 years ago

A lot of music library managers (including good old iTunes) let you add a lot of various and esoteric song tag data as columns in the track list. The problem was always that the column arrangement that makes sense for one style of music doesn't make sense for another.

One missing feature that I've never seen in a music library application is the ability to have "presets" for these column headers that you can easily switch between. So you can turn on the "classical" preset which shows Composer, Performer, etc., and you can turn on the "contemporary" preset which shows Artist and Album.

Semaphor• 4 years ago

That’s actually something MediaMonkey [0] (Windows-only) had for a long time (I assume the current electron version as well, I use the last native version). Classical is one of the defaults, but you can add whatever.

It’s one of my favorite pieces of software, a true music library manager and not a music player with some management features tacked on.

[0] https://www.mediamonkey.com/

Aidevah• 4 years ago

Agreed, I've tested many self-hosted music streaming servers and decided to settle with navidrome in the end but not without major reservations. I personally don't mind sorting music via album/song, but the lack of composer field is really painful. Unfortunately the omission of composer information occurs in the subsonic api, which have a large number of compatible players, so adding the field is not as straightforward if one wishes to do it in an interoperable way.

I used to be quite happy with a local music library and playing it with mpd, but at some point I got sick of maintaining over 100 GB of music library across multiple devices. The streaming solutions for mpd are quite awkward to use, otherwise I would have stuck with it just to get the composer field.

thaumasiotes• 4 years ago

There's an analogous problem going on on musescore.com, where you're supposed to list the work your score is based on. Viewing a score on the site will then include links to other scores based on the same work.

I uploaded several pieces from Swan Lake, and listed each of them as the particular piece that they were. For example, https://musescore.com/user/36584999/scores/6642212 is Swan Lake, No. 4 (Pas de trois), part IV.

But every time I list a piece like this, it gets "corrected" to the canonical reference, which is "Swan Lake". And the "other scores based on the same work" are totally useless, because they can be any piece from anywhere in the ballet. It's more than 600 pages long! Where's the value add from removing my indication of exactly which piece of Swan Lake I just scored?

TheOtherHobbes• 4 years ago

It's a sleeve notes problem. Ideally you'd want a complete copy of all the info in the sleeve/inlay. This includes some combination of headline writer/composer, producer, conductor, headline performer, co-writers, arrangers, session players, engineer, studio, lyrics, and random sleeve notes/essay, possibly in multiple languages.

Per track.

Plus artwork. With credits.

Plus random thank-yous and special thank-yous to sponsors and other contributors.

The importance of each field varies with each project. For classical conductor showcase projects the conductor is the headliner. For classical composers, obviously it's the composer. For compilations it can be a concept like "Classical Chill" or "Early Music from the 15th Century." For pop projects it's usually the band/artist name. For movie tie-ins, it's the movie, or perhaps the franchise.

And so on.

You get all of this for free on CD/vinyl labels, and the fact that it's freeform doesn't matter because the sleeve artwork and visual placing of the elements make it clear who/what matters.

Trying to fit this into a single fixed metadata schema is madness. It can't be done.

This is one of those situations where trying to data-fy the information leads to huge loss of detail. You get some nominal (poor) searchability, but you give up a huge amount of interesting supporting information.

What the industry needs is a digital sleeve note standard which can handle all of this and more.

But metadata is designed for sales and stream tracking, not for user satisfaction. Which is why it's so bad at capturing what users want.

nailer• 4 years ago

The odd one for me is Pendulum - there’s two different Australian acts, both electronic, with the name Pendulum. For some reason the newer Pendulum really liked the name and weren’t bothered / ethically concerned that the Pendulum that made ‘Coma’ had already existed. As of today Spotify has songs from both artists under the same artist name.

lukeh• 4 years ago

Plenty of duplicate artist names, Spotify artist support will split them into separate artists if you contact them, and now aggregators like DistroKid let you specify the actual artist ID when uploading (at least for some DSPs). It’s just another instance of there are only two hard problems :)

tomcam4• 4 years ago

Yeah my favorite is definitely Pendulum.

lukaslalinsky• 4 years ago

There are much worse things than Pendulum. John Williams seems a very popular name for artists. :) Back when I was designing the "new" MusicBrainz database, that's what we have been using as the test case for this.

nescioquid2• 4 years ago

I wonder what the "You suck, Flying Circus" sounds like.

Bud• 4 years ago

Pretty disappointing to see an entire article like this go by without a single mention of classical music. Also known as basically all music prior to 1940 plus a lot of the most enduring music since.

Karuma• 4 years ago

They do mention Tchaikovsky there... Or what do you mean, exactly?

tomcam4• 4 years ago

Not sure I understand? There are accepted names/sequencing for all major Baroque, Classical, Romantic etc. artists: Mozart K numbers, Bach BWV numbers, etc. And usually there’s a close to unique name like Adagio in D Minor, Piano Sonata in Bb Major, etc.

repsilat• 4 years ago

I guess they mean that there are no original recordings for most of it, lots of cover bands etc.

zokier• 4 years ago

Also known as basically all music prior to 1940

Sure, if you ignore blues&jazz, traditional folk, and all non-western music.

eyelidlessness2• 4 years ago

I was all geared up for a “misconceptions software developers have about _” post but this seems mostly like … Unicode exists and data modeling should at least comprehend the domain?

jandrese• 4 years ago

Or really: Natural Language Processing still sucks. Trying to fit anything human produced into a reasonable data model is doomed to failure.

recursive• 4 years ago

I was hoping this was going to be about the musical part of music rather than the book keeping.

cratermoon• 4 years ago

I was hoping it was going to be about music engraving. See e.g. https://twitter.com/ThreatNotation

cosmotic• 4 years ago

Very "creative"

tomcam4• 4 years ago

They missed a few common ones; Led Zeppelin’s fourth album with Stairway to Heaven on it was named only with Prince-like symbols, and of course the group named The The (which Google indexes well).

deadbyte• 4 years ago

The Inertia Variations is fantastically uninspiring.

cgh• 4 years ago

Both of these are mentioned in the article, or were you referring to something else?

sriku2• 4 years ago

I understand "horrible" in this case in a somewhat lighthearted sense. But it points to a tendency to simplify cultural complexity in the interest of "legibility" from the system perspective. The same diversity of cases need to be accommodated in much design done for, say, a diverse country like India where though urban populace me be English savvy, rural populace aren't necessarily so ... And even in urban areas local language and culture dominates.

A place this diversity crops up most commonly is the so called "standard" firstname middlename lastname fields. This just doesn't work in India where people may have even 7 components to their names ... and who are you, dear system developer, to dictate what their name should be like.

The name structure diversity I understand is a problem for Europe too with hyphenated names. This brilliant sketch by Fry and Laurie says it all - https://youtu.be/1LopIroSjsU

wincy• 4 years ago

I mean, we get to dictate it because we’re writing the code and don’t know? I guess enterprising Indian developers should help it get fixed. As the whole world gets online this stuff gets sorted out, eventually. It’s not as if it’s malice or something. There are Americans with longer names though, I had a friend whose full name was five words, and all of his sisters and brothers also had “three middle names”.

sriku2• 4 years ago

Not malice ... rather incompetence. If we can't handle this much accessibility, what hope do we have of making "ethical AI"?

defensem3ch• 4 years ago

There's also Goto80's Files in Space https://www.goto80.com/goto80-files-in-space-mcmp3-data-airl...

nescioquid2• 4 years ago

Here "edge-case" indicates that music is not mere noise, while somehow remaining an instance of "marketing" or "acculturation".

This is a good example of what happens when the insensate take power and pander to those who just want to read the "rule book".

No art. No expression. No bother.

Let's just agree on some fucking fiction that gets record label to bother with classical music meta-data in a way that meets the real world. Can we start there, at least?

edit: click the inverted triangle if you are a jack-ass

thewakalix• 4 years ago

You didn’t say “only if”, so it’s still possible to click the inverted triangle without being a jack-ass.

lacrosse_tannin• 4 years ago

There's also different releases of the same album... 2020 remaster, vinyl rip, web flac, japan edition...

Sometimes the music app just smooshes all together so the "album" looks like it has 4 of each track

thristian• 4 years ago

Peter Gabriel's first four solo albums were untitled. Well, the fourth was titled in America, but untitled everywhere else. Fans tend to distinguish them by the cover art: there's "Car", "Scratch", "Melt", and "Security".

sharph• 4 years ago

It's worse than that, they were self-titled. So if there were some mechanism for numbering untitled albums, it can break down.

dehrmann• 4 years ago

Weezer's self-titled albums are distinguished by background color of the cover art.

kingcharles4• 4 years ago

LOL. I used to work for PG. I created a schema for storing music metadata for him, so it's somewhat ironic.

eashman2• 4 years ago

Oh please tell this old Genesis fan some more. Gabriel was a very forward thinking musician from a technology perspective. See this clip for his demonstration of an early sampler [0]

[0] https://youtu.be/7Xfj5n1kYXY

egypturnash• 4 years ago

I am amused that despite mentioning both Keygen Church and Master Boot Record, this list refrains from mentioning that the former is a side project of the latter.

And, yeah, if I was still collecting my music as CDs? Keygen Church would be filed next to the MBR because of that, not over in the Ks.

Lammy• 4 years ago

Most albums have several versions/releases

And some times they're altered without any obviously-visible indication like differing number of tracks or "edition" naming, e.g.:

- MM..FOOD? (2004) [RSE0051-2] where "Kookies" samples Sesame Street: https://www.youtube.com/watch?v=4RYCLfGE-_Q

- MM..FOOD? (2007) [RSE0084-2] where "Kookies" is totally re-recorded with a different and non-infringing beat: https://www.youtube.com/watch?v=8iYSwvdEfeY

If you use a streaming service you're hearing the latter and might not ever realize the former existed. The re-record is a fine song, but the original fits much better with the rest of the album.

dotancohen5• 4 years ago

This is my problem with Rust in Peace. It was a terrific album, but Dave didn't want to continue paying royalties to people he no longer likes. So in 2004 or so almost the whole album was rerecorded with studio musicians. It is horrible, mechanical, flat. Even some of the vocals were redone, poorly.

I'll sign up for the streaming service that provides the original 1990 Rust in Peace.

scns3• 4 years ago

Is that the album where they blew the advance on drugs and had none left for good mixing and mastering?

slothtrop3• 4 years ago

Yeah that was particularly obnoxious. According to discogs.com I think there are more recent reissues/remasters of the original version.

kzrdude3• 4 years ago

This annoys me in Spotify so much, that we often only get one version of the album, and usually a new one instead of the original.

And maybe I should feel like it's a luxury to have extra bonus tracks with more takes at the end - but I lean over to the album purity side more - the album is one artwork by itself, I want the original track listing and nothing more, as "the album".

iggldiggl• 4 years ago

If you use a streaming service you're hearing the latter and might not ever realize the former existed.

On a number of live Bob Dylan albums, for inexplicable reasons all online versions (not just streaming, but the download releaes, too) are now missing the between-tracks content that on the physical CDs is part of the pregap, which means you're missing out on the funny stage banter on the 1964 Halloween concert, or the infamous heckling on the electric half of the 1966 Manchester concert.

thrdbndndn2• 4 years ago

Weird album names:
† from Justice
from David Bowie
' from Frank Zappa
() from Sigur Rós

How are these albums names "edge cases"? I feel like ANY software would deal with them just fine. Hell, the last two don't even use non-ascii characters (though you definitely should have unicode support regardless).

Edit: I just realize HN won't the star symbol (U+2605). Which is weird since it does support characters like Chinese/Japanese glyphs あいうえお你好

Beltalowda2• 4 years ago

There are actually some challenges with stuff like that; how do you sort "'" for example? Most people call the album "Apostrophe"; should it be sorted under "A"? What about the "†"? "D" for dagger? Somewhere else? How do you find things like that in the search box? Should searching for "star" return that Bowie album?

Edit: I just realize HN can't display [star]

It doesn't do emojis and various graphic characters. This is supposedly a "feature" shrug-emoji-here

Freak_NL• 4 years ago

Sorting is usually solved by allowing for optional separate sorting-key tags. So ' :

    album='
    albumsort=Apostrophe

These can act as search input as well, so searching for 'apostrophe' could yield it depending on the software used.

† can either just sort wherever Unicode puts it, or if the album has a known pronunciation like 'star' above, have a sorting key too.

kzrdude3• 4 years ago

The aversion to emojis is strange. I've come around to thinking of them as a net positive. Just text is sterile and the tone often comes off wrong! Smiles lighten up conversation, we need that to not lose track of each other's humanity when arguing in comment threads. :)

kingcharles4• 4 years ago

HN strips emojis. I can't even paste my web site domain on here.

kingcharles4• 4 years ago

This is more of an edge case:

https://cdn.shopify.com/s/files/1/0022/8193/0805/products/im...

When music downloads were first starting up in the late 90s I was responsible for building a schema to store all the world's music. We had to rip millions of CDs. When I got that Aphex Twin one above I gave up. I think we just called it [Track 2] or [Maths symbols] or something.

maxbendick2• 4 years ago

My time to shine! I work in music distribution.

The Apple Music style guide is a great and accessible place to look if you're interested in music metadata: https://help.apple.com/itc/musicstyleguide/en.lproj/static.h...

DDEX is the set of metadata standards used across the music industry: https://ddex.net/

Also worth knowing that not all content on streaming services is music! Some of it is spoken word, ASMR, non-music field recordings, etc. The difference between sound and music is subjective of course though.

On an artistic note, music can of course be presented in an infinite amount of ways. Not all these can be represented (i.e., re-presented) on streaming platforms. Installations and generative music for example. That's ok! To be able to represent all this music (and non-music) requires restrictions, otherwise we wouldn't be able to create programs for it in the first place.

Sometimes restrictions on sound content are to make the streaming services friendlier to use, like the formatting of featured artists in titles. Royalty laws put restrictions on the handling of different roles, like songwriters and composers.

BUT nothing is stopping you from building your own way of representing music digitally, so long as you follow relevant laws.

infofarmer2• 4 years ago

Pleasant to see Apple has made some progress on their metadata guidelines (it used to be abhorrent for a company saved by the iPod). Looks like a useful one-pager to get a quick overview of the challenges without getting into any of them.

For a proper attempt to preserve metadata, I'd refer to the MusicBrainz Style Guideline — https://musicbrainz.org/doc/Style

Thorrez• 4 years ago

Sometimes restrictions on sound content

Is there some sort of rule against sirens?

The song "Car Alarm" prominently features a car alarm on Youtube, but not on Spotify. This could just be a choice by the artists though, rather than a platform restriction.

https://www.youtube.com/watch?v=xV7nHX2RLjQ

https://open.spotify.com/album/2u4HDb57v96iiJZUC7PqOx?highli...

spicyjpeg• 4 years ago

I have noticed that it's fairly common for many songs to feature random car sounds in their YouTube mixes, but not in the final masters that end up on iTunes and streaming services. It is also relatively common to have "watermarks" (e.g. additional short vocal clips at the beginning and end) added to the YT/Soundcloud mix, presumably to make it easier to figure out if a DJ just ripped the song off YT.

withinboredom• 4 years ago

Probably not a rule, but maybe more like a liability thing? If people listen to your music while driving, you don’t want them to freak out from the sound effects and then sue you because they got in a car accident.

ratww• 4 years ago

That's a music video. They traditionally have sounds and dialog that aren't part of the music itself.

Here's the regular version, same as Spotify, also on Youtube: https://www.youtube.com/watch?v=N4inWiKCt-E

I don't know if Spotify has music videos, but Apple Music for example does, so you can have both versions.

seba_dos13• 4 years ago

Also worth knowing that not all content on streaming services is music! Some of it is spoken word, ASMR, non-music field recordings, etc. The difference between sound and music is subjective of course though.

Even some albums that are clearly about music can contain non-music, take Tenacious D's skits as an example.

fivre• 4 years ago

Maybe you can explain one of my long-burning questions: why are non-English tracks handled so inconsistently? I'll sometimes see an English translation, sometimes a transliteration (for non-Latin scripts), and sometimes the original language. It's not consistent which gets used even for a single artist, and sometimes changes to a different one.

The albums Ангедония (https://open.spotify.com/album/1HxmgR8wpc1ySplYCTNwaW or https://i.imgur.com/ndOXsgL.png) and Продано! (https://open.spotify.com/album/5kp7j9B4TDA3VfhaXz9XcJ or https://i.imgur.com/QJNjXc1.png), for example, both contain a version of the track "Рижская". The former uses the original Russian while the latter uses Rigas', which is a sort-of translation (it's an adjectival version of "Riga", which doesn't really exist in English).

https://help.apple.com/itc/musicstyleguide/en.lproj/static.h... indicates that there are separate fields for each of the original, the translation, and the transliteration, but it doesn't seem like anyone actually _uses_ these, and instead just picks one arbitrarily to stuff into the main/native field. I'm not sure if Spotify even has a way to display the alternates--I can't recall ever seeing a toggle for it, though idk if I've ever actually encountered a track that includes the variants, since you can't see raw metadata on Spotify.

Is it just that music labels do a lackluster job of handling metadata? I expect this is probably the case, since classical music is similarly messy--it's a crapshoot whether the composer or performer gets used as the artist. For that I've at least seen the composer metadata field populated sometimes, but internationalization fields don't seem to be used ever.

stavros5• 4 years ago

It's really hard finding Greek songs on Spotify for this reason. There are many ways to transliterate (eg "ευχή" can be "efhi", "euxi", and all the combinations thereof, and that's just with two ambiguous letters) and no names are in Greek, so it's basically a game of trying to guess the transliteration for the track you want.

bobuk• 4 years ago

quick upvote for examples from Yanka. I thought that I'm the only one who still listening her.

maxbendick2• 4 years ago

This is an area that many parties handle differently. Streaming services, artists, labels, and distributors don't all handle translation metadata consistently. There are specs in DDEX for this, but it's a matter of support and doing the translation work AFAIK.

slaymaker1907• 4 years ago

I still think it's pretty bad for classical music. An important question for various pieces is who plays a particular instrument, but there really isn't a nice way to encode this in the metadata (encoding the relation of both person and role where role could be anything).

I also usually want to know who the conductor was and what was the orchestra

azalemeth• 4 years ago

This, many times over. Id3 tags are one thing but all decent classical record labels (most of whom have always been drm-free) like Hyperion or Naxos give you a pdf cd jewel case booklet. I've got recordings of many friends and it's always a pleasure to see their names in this. I don't use streaming services and one of the unexpected joys of iTunes is that the album art can contain a pdf with a cover image. The rest of the UI mostly assumes that you're listening to non-classical music though. I'd love to able to search for e.g. "Victoria's Requiem" and get albums with sensible names, even if they're in Latin, seemlessly grouped by performance.

eliaspro2• 4 years ago

A few years ago, we started working on a platform for classical music [1] and took great pride in our metadata and our underlying schema which basically covered 99% of all the weird edge cases classical music has.

Unfortunately, we couldn't make it work economically even though we tried several approaches for monetization and had to shutdown a while ago.

[1] https://grammofy.com/

inopinatus2• 4 years ago

I'd expect publishers of classical music to know the difference between a composer, a conductor, a soloist, and an orchestra, but no, apparently not. They freely mash them up across fields. The difference between a symphony and an album is often mangled too, especially on releases containing multiple works. And for fuck's sake, no, a movement is not a song. I don't play, "like", flag, list, cross-fade, or reshuffle them independently.

Music metadata has been optimized to suit the marketing preferences of contemporary record labels. But more general classes of music won't shift units on Spotify so the message is loud and clear, a resounding "get fucked" from the assholes that run this industry.

(One side of my family is in the recording business. I can confirm the whole thing is run by assholes.)

gvurrdon• 4 years ago

This is also important for jazz. Should I wish, for example, to find albums where a particular bassist is playing then this would be rather difficult unless they happen to be the leader for that particular recording. Also, it would be particularly helpful to have subgenres, e.g. jazz -> post bop, jazz -> free, etc.

infofarmer2• 4 years ago

There's a reason why MusicBrainz SQL schema looks like what it looks like, after years of careful simplification to achieve an optimum approximation of reality.

https://musicbrainz.org/doc/MusicBrainz_Database/Schema

10+ years ago at Zvooq, during the golden age of music metadata and recsys startups, we tried to really solve entity resolution in the domain. I've never seen another streaming service make an honest attempt at this, even at Spotify with their 2014 acquihire of the Echo Nest. You still get Utada and Hikaru Utada as ~completely separate entities

willhinsa• 4 years ago

what.cd had a really solid model as a user, but I've certainly been pleasantly pleased with musicbrainz.org

throwaway483752• 4 years ago

I really miss what.cd

dewey• 4 years ago

What.cd had a very simple data model and it had the same issue as many music sites trying to fit it into a simple schema (and not a more sophisticated one like MB). The first issue that comes to mind was artist with the same name, these ended up being the same artist (Just like on Last.fm) if not split by hand by changing the artist name.

The schema is here in case you are curious: https://github.com/WhatCD/Gazelle/blob/master/gazelle.sql

Helmut10001• 4 years ago

I think MusicBrainz did a great job to find some common agreed denominator between all these styles and formats for music metadata. After setting up Funkwhale [1], I went through all my 16000 MP3s and synced those with the MusicBrainz Library, also adding new entries. It went pretty smooth and my library looks much better now. It is also really fun to listen again, when everything is structured and labeled.

(btw., yes, I bought a large percentage and, nowadays, mostly buy on bandcamp [2]).

[1] https://funkwhale.audio/ [2]: https://bandcamp.com/

andybak• 4 years ago

Ha! I contributed the initial classical style guide many, many moons ago.

I'm still technically an automod although I haven't participated for a long time.

rcthompson• 4 years ago

AC/DC is another edge case I've seen, mainly because in some contexts slashes are used as separators between multiple artists, so "AC/DC" could be misinterpreted as a list of 2 artists named "AC" and "DC".

amelius• 4 years ago

Doesn't work well in filenames either.

SAI_Peregrinus2• 4 years ago

"/Τĥιs ñåmè įß ą váĺîδ POSIX paτĥ" is my display name on some services at my job. It helps to remind people to consider things like spaces, homoglyph attacks, and non-ASCII characters.

mehrzad2• 4 years ago

A funny edge case I've found in one of the most popular music services (Last.fm). There is an instrumental post-rock ish band called Collections of Colonies of Bees and they have an album where almost all the tracks have the same name; they are all titled "Fun" except for the last track titled "Funeral". https://open.spotify.com/album/0LP8DTy4amlLIVXGOjvCo2?si=HJ3...

This has caused last.fm to think "Fun" is one of my most played songs of the year even though they were all separate tracks.

aasasd3• 4 years ago

I bet the most popular track name in techno is “Untitled”.

mehrzad2• 4 years ago

I would guess that the problem only appears if it's the same artist and same album ID. Not sure if techno artists use "Untitled" multiple times on the same album, haha.

zimmund• 4 years ago

Last.fm doesn't allow "untitled", "various artists", "track 01" and other titles that may suggest the song is incorrectly tagged. This leads, of course, to edge cases where an artist has a track named "untitled" and it can't be scrobbled.

Moreover, last.fm also has an issue with unicode characters and some artists/albums may have two or more different profiles. It's a mess!

Tokkemon2• 4 years ago

Most of this is great, but I'm calling BS on the Tchaikovsky example. Yes, there's been multiple historical spellings of his name, and to be absolutely accurate you would use the Cyrillic. However, the vast, vast majority of use cases should use the modern accepted spelling in English when referring to Tchaikovsky, which is what Wikipedia uses. There's a line you have to draw somewhere between accuracy and comprehension by the user and "Pyotr Ilyich Tchaikovsky" is a pretty damn good one. The only other one I see commonly these days is the slightly more Anglicized "Peter Ilyich Tchaikovsky" but that's it. Most other variants are historical or just have different editorial decisions, but they all refer back to the same guy, not a living artist who is changing his stage name all the time.

blahedo• 4 years ago

...in English...

Even in English-language contexts I've seen variants of the German-style transliteration on occasion. But more importantly, the author is not restricting himself to English, and indeed the musicbrainz link gets to such a high count precisely by including other languages' transliterations. Only three are listed as English, and they're the two you mention (with/without Anglo first name) plus one that omits the middle name.

Were you saying that all other languages should use "the modern accepted spelling in English", or were you just ignoring the fact that speakers of other languages (or owners of foreign-published albums) might use music databases?

kzrdude3• 4 years ago

The database might implicitly force a monolingual perspective on the dataset if they can't handle multiple translations of several fields. It's not an easy problem to manage, I imagine.

xenadu02• 4 years ago

Another famous one is Prince's "Love Symbol no 2".

Fun fact: he adopted it as his stage name to spite Warner Brothers Music, due to the unfavorable contract terms and the creative meddling they kept doing (among other things). By terms of the contract they had to go along with whatever stage name he picked. They had to spend a bunch of money having a custom font created, then mailed on floppy disk to every music reviewer, newspaper, magazine, radio station, etc as part of album promotion. No one was on the internet in those days so it ended up causing them a lot of trouble. I guess he found his own loophole in the contract.

buro9• 4 years ago

I've recently wondered whether it's worth periodically reprocessing the metadata on music files with musicbrainz Picard.

It was all done when added to the catalog, but presumably the metadata could be updated and a periodic refresh would be a benefit.

can16358p• 4 years ago

What is the deal with Japanese exclusive bonus tracks/promo thing?

Interested about why a significant number of albums have Japanese promo versions with extra tracks. Is that a cultural thing about Japan or is there another specific reason? (I know it's not only Japan but mostly Japanese editions that I see throughout the web)

As if, almost even brouillard has a Japanese bonus track named brouillard in their album brouillard.

Hamuko2• 4 years ago

I heard somewhere that it was an incentive for the Japanese market, which still today is a large physical CD market, to buy the locally distributed releases of albums instead of just buying the original foreign release, which may have been available earlier and for cheaper, or if they already bought the original release, buy it again.

dotancohen5• 4 years ago

Is this why the super-high quality vinyls were also available only in Japan? I cannot remember the name of the company, but there is (was) some company selling increased dynamic range recordings of popular albums, on very durable (as in, regularly replace the needle) vinyl. I had heard an imported (from Japan) Dark Side of the Moon recording on that once, and it just blew me away.

scns3• 4 years ago

I once read somehere that CDs are expensive when new, can't confirm though. When i was there i bought them used for cheap. Slayers' Hell Awaits, colored like egg yolk ~6€ IIRC, thank you very much.

whyoh2• 4 years ago

Tagging challenges are one reason why I sort and browse music by folders/filesystem.

klez• 4 years ago

Aren't you just moving the complexity from tags to the filesystem that way? How does that solve the problem in a way that tags don't?

whyoh2• 4 years ago

Yeah I should have made it clear that I mean specific problems, like musicians who release an album under a different name... With a folder based approach it's easier to put all of those in one place. The filesystem also has some drawbacks, like a more limited character set. But all in all, I like it.

throwaway384813• 4 years ago

Bookmarking this for future reference!

I recently started a fun/side project which aims to improve my personal tedious process of (a) managing a queue of things I want to listen to, (b) keeping track of which of them I listened/liked and (c) easily sharing my "discovery activity" with my friends who share the same passion.

The idea is that you can add items to a "Queue" list, using Discogs Master URLs (e.g. https://www.discogs.com/master/1994809-Alt%C4%B1n-G%C3%BCn-Y...) or Spotify album links, because this is what I use in my day to day.

So I've started by parsing Discogs' data dumps[1] to backfill my database (aka. "release groups" in MusicBrainz), so that I avoid hitting Discogs API with every user query) with all "master" releases. But then realized that (a) they don't include cover images and (b) not every release has a corresponding "master" release and (c) they're not updated frequently enough (maybe monthly).

Then I thought of using MusicBrainz data dumps[2] to augment the entries by Discogs in my database, which do provide cover art images, but then how do I correlate them with the already-inserted Discogs releases? Fortunately there's a Discogs URL for many release groups from MusicBrainz; unfortunately only 50% of them do have it.

Then I could perhaps use solely MusicBrainz data and not Discogs at all, but then what if people are mostly using Discogs links? This will result in two identical records on my database, pointing to the same underlying release but in different sites (Discogs & MusicBrainz). Perhaps this can only be solved by human moderation and providing the ability to "merge two items".

Then I thought of actually contributing to MusicBrainz and add the Discogs URL for every release group.

Another interesting assumption I made, which turned out incorrect, was the every [artist, release title] pair was unique. But that's not the case - an artist may have multiple "master" releases (i.e. completely different tracks inside) with the same title.

[1] http://data.discogs.com/

[2] https://musicbrainz.org/doc/MusicBrainz_Database

creshal• 4 years ago

- Artist names can become rather long too, especially for collaborations. Some anime OSTs easily break 256 characters.

- And for song titles that need careful escaping, QuelI->EX[cez]->{kranz}; and EXEC_SOL=FAGE/. have tripped up more than one CLI music player in my experience.

morelisp• 4 years ago

A simpler explanation is that your computer has an unstable grathnode installed.

photon-torpedo• 4 years ago

To add to the list: Song titles that cannot be fully represented in Unicode. Example (also to plug my favorite band): Magma, a French progressive rock [*] band that's been (on and off) active for more than 50 years, mostly uses their own invented language "Kobaïan" where some of the characters use diacritics that are (to the best of my knowledge) not present in Unicode. E.g. there's an S with a kind of crown on top.

[*] Actually their style of music is unique enough that it's given rise to its own genre, "Zeuhl".

gorgoiler• 4 years ago

Catalogue numbers!

You wouldn’t index a collection of records by the shape of its sonic wave forms, would you?

No more would you index by the artistic representations on the cover which, as we have seen, includes everything including the words purporting to be the artist name and the title of the release.

No, of course not! You use the barcode on the back!

https://en.m.wikipedia.org/wiki/Catalog_number_(music)

For white labels, I guess you have to hide them in a pile until they get a release.

shiomiru• 4 years ago

Unicode support is one thing. I frequently have to re-tag Japanese music because its tags are encoded in shift-jis which in general doesn't work on Linux.

eschneider• 4 years ago

And for new bands: Maybe try not to name yourself after a Siri/Alexa/Whatever revered word. Jus try asking Siri to play a song by "Them." Ugh.

rzzzt2• 4 years ago

"The The" is my go-to example!

tinus_hn• 4 years ago

It’s a very ingenious and effective anti-piracy measure!

FabHK• 4 years ago

Meta Comment:

I much prefer this informative post (with a clear title and plenty of examples and explanations) to the uncommented "Falsehoods programmers believe about ..." listicles.

MereInterest• 4 years ago

Absolutely. Stating that an assumption is false with no further information doesn't help me update my mental model of the world. Giving an example of how an assumption is incorrect shows the type of errors that can occur, when they can occur, and lets the reader figure out a better way to work around the false assumption.

DonaldFisk• 4 years ago

I've noticed some of these for years. When designing a music database, you do need to be aware of them. You need to use unique identifiers instead of names. last.fm doesn't do this.

There were four bands called Kaleidoscope, all active around the same time: one from England, one from Mexico, one from the USA, and one from Canada. To disambiguate them, generally the country is specified, e.g. Kaleidoscope (UK).

There were two bands called Bulldog Breed, both of which recorded albums called Made in England. They're decades apart, but as albums can be re-released, they're normally disambiguated by record label, e.g. Bulldog Breed: Made in England (Deram).

There are two Nirvanas, the original UK-based band, and the US one. The UK Nirvana recorded a cover of Lithium.

You'll be aware that the Beatles recorded an album called Revolver. A Russian band, called Revolver, recorded an album called Beatles.

Do you include the definite article in the band's name? If so, what about The The?

Tintern Abbey released a single single, the A-side of which was called Beeside.

mauvehaus• 4 years ago

I've always thought it'd be a bit cheeky to start a band and name the first three albums "Self-Titled", "Eponymous"[0], and give the third album the name of the band.

Fortunately, I lack any amount of musical talent, so this will never happen.

[0] I know, REM beat me to it.

cpeterso• 4 years ago

Do you include the definite article in the band's name? If so, what about The The?

My SiriusXM radio omits leading articles like “The”, so “The Beatles” is displayed as “Beatles”. I alway long wondered how it would display “The The”. Would it be “The”, empty string “”, “The The”, or something else entirely. I eventually learned the answer: “The The”. Which is good, but then made me frustrated that it wouldn’t allow the “The Beatles” to be “The Beatles”.

vermooten• 4 years ago

Don’t forget how much streaming services struggle with classical music meta data.

philjohn• 4 years ago

Similar problems when dealing with books.

J.K. Rowling has written works under other names, e.g. Newt Scamander.

And then you have the case of the (actually recorded this way by the Library of Congress) author "Mark Twain (ghost, through mediumship)"

moogly2• 4 years ago

Adding to the list:

There are two well-known Avishai Cohens, both jazz musicians (so you can't disambiguate on the broader genre), both from Israel (so you can't disambiguate on country of origin), but luckily they do play different instruments (bass and trumpet).

---

The Australian post-rock/shoegazeish band The Dead Sea https://www.discogs.com/artist/1450625-The-Dead-Sea released a self-titled EP with 3 tracks in 2006.

2009 they released a full self-titled album (only 1 song from the EP featured on it).

I can't even find the "second" release on discogs.

Both were self-released, so no way to disambiguate on ISMN or record label/publisher.

So you'd have to disambiguate only on the year of release, or define what demarcates an EP from a full album and mark them as such (there are some competing definitions out there, since it became less clear after the introduction of the CD, and now digital, but even back then you had complections like maxi singles and mini-LPs, IIRC).

---

Many of John Zorn's releases tend to be headaches:

"Tap"

Composed and produced by: John Zorn

Performed by: Pat Metheny

Subtitle "Book of Angels, vol. 20"

Overarching series "Masada Book 2"

Released on Tzadik (Zorn's personal label) as a John Zorn/Pat Metheny output.

Licensed to Nonesuch (Metheny's label) as a Pat Metheny release with different cover.

slothtrop3• 4 years ago

You also can't find Zorn's music streaming anywhere. I've heard of one store that offered digital purchases, but not for the whole discography.

You can fill a large closet collecting just his stuff so I'm not sure why digital is off the table.

adrianmonk• 4 years ago

There are two well-known Avishai Cohens, both jazz musicians (so you can't disambiguate on the broader genre), both from Israel (so you can't disambiguate on country of origin), but luckily they do play different instruments (bass and trumpet).

Along very similar lines, there are also famously two musicians named Bill Evans. Both play jazz (genre) and both are from the US (country). As a bonus, both have been members of bands led by Miles Davis. And again you can differentiate them by instrument (piano and saxophone).

Which is sort of what Wikipedia article titles do to differentiate them:

https://en.wikipedia.org/wiki/Bill_Evans

https://en.wikipedia.org/wiki/Bill_Evans_%28saxophonist%29

I guess Wikipedia also differentiates by level of fame, since the pianist article title doesn't have a qualifier.

parenthesis• 4 years ago

I was searching for the artist Urban Soul on Tidal to discover there's another artist called Urban Soul (I wanted the Roland Clark project) with albums by both mixed up together in the search result.

'Twas all in vain, since I wanted a particular remix of `Alright' — which I have on a vinyl promo less that a metre from where I am sitting (I can remember clear as day buying it more than 25 years ago) — which Tidal don't have. So I ended up listening to it on YouTube.

inetsee• 4 years ago

How about the horrible edge case when trying to read this pale blue text against a white background. It's been a while since I criticized the readability of a web page, but I think this page deserves a critique.

SAI_Peregrinus2• 4 years ago

One of the simplest data model mistakes that seems nearly universal in music software is pretending that some things are single values when they're really lists.

Take "artist". An album can have a compilation of tracks, each from a different artist. A given track can have multiple artists, eg a guest singer. There can be multiple composers of a track, there can be multiple writers of lyrics, and there are usually several performers (even if they all are part of the same band).

An album can have multiple names, or no name at all (which often means people refer to it in different ways, giving it multiple effective names). Likewise with tracks.

Albums and tracks get re-released, sometimes re-mastered, sometimes re-recorded entirely, without the names changing.

The same recording of a given track might be released on multiple albums, eg as a single in a pre-release and then within the main album.

Tracks might be mastered differently for different media (Radio vs CD vs background use in cinema, etc).

Names of any of these things don't have to be representable in any existing text encoding, or even any combination of encodings.

Artist name should be a list (or other multiple element structure).

Track name should be a list.

Album name should be a list.

Etc. Assume a malicious artist will find a way to make the assumption that any field has a unique value incorrect.

wolverine8764• 4 years ago

or other multiple element structure

Right. It's a relational database with lots of many-to-many relationships. We need to stop trying to make it a flat file. A FOSS SQLite schema that everyone can embed, or even a turnkey little db, would be wonderful.

wdr1• 4 years ago

I feel like musicians were doing Little Bobby Drop Tables long before the XKCD comic.

joshspankit3• 4 years ago

100%

joe__f2• 4 years ago

Does anyone recognise the equation from the Aphex Twin track?

∆Mᵢ⁻¹=−α ∑ Dᵢ[η][ ∑ Fjᵢ[η−1]+Fextᵢ [η⁻¹]]

With the ∆M and the sum over Fs, I wondered if it might be something to do with rocket dynamics derived from Newton's second law?

trelliscoded• 4 years ago

It's pseudomathematical nonsense. The Fji term might be a hint about the image hidden in the track when plotted on a spectrogram, but the rest of it doesn't make much sense. You can see the original title here:

https://i.discogs.com/YC8CLuli-NW3VB5z7eA9mALjnL2Qef4aGyDYH4...

As an example, why would the index of the last term be inverted like that? Maybe it's supposed to mean something phonetically, but I'm not a music person so I can't figure it out.

joe__f2• 4 years ago

Oh I didn't realise the eta was the index, I was reading the sum as over i on the right hand side

hamstergene• 4 years ago

Electronic music is also full of remixes, for example:

Artist: Moby

Title: Simple Love (B-Motion Remix)

The actual artist is B-Motion and any kind of browser/catalog must group this track with other B-Motion music, or in worst case both, because remixes are in entirely different genres than the original.

So "Artist" is not necessarily the artist.

ux• 4 years ago

I wrote something similar a while back: http://blog.pkh.me/p/15-the-music-classifying-nightmare.html

joshspankit3• 4 years ago

It's worth noting as well, that some of these edge cases are directly because of artists intentionally "messing with" whatever they perceive as hard rules.

The reason I mention this is because you cannot simply look at a list of all edge cases, identify what schema you can use to encapsulate them all, and then build your service around the idea that the schema will never change.

As a very simple (I hope contrived) example: You could decide to use a string delimiter so that you can capture "all the strings" and still use string interpolation. You then look through all the edge cases and see "No one has named anything with U+1F7F6, so we're safe to use that". 100% guarantee that as soon as someone sees that they will name a track or an album in the exact way they need to to break your system.

kazinator• 4 years ago

I don't think it's meaningful to talk about edge cases without a set of requirements connecting inputs to desired outputs: what you want to do with the data, or hope to get out of it.

Features like long album names, empty names, or names with international characters from Unicode, are not hardly edge cases. Edge cases have to have some surprise to them.

Coming from a blank slate, you'd never approach a problem with the assumption that all strings are 32 characters or less, confined to 7 bit ASCII, and are never empty and then paint your face surprised call "edge case!" when they aren't.

Horrible edge cases when dealing with music

Related Stories