Hacker News new | ask | show | jobs
by maxbendick 1531 days 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.

5 comments

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

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.)

I like being able to address and like/favorite individual movements. Beethoven Piano Sonata #14 (“Moonlight”) is a masterpiece, but the second movement is <expletive>. I don’t know if I could listen to it hundreds of times like I have the first and third movements.
Many classical works have more than one 'track'. Musicians, critics, etc. usually refer to them by the tempo marking text. A lot of music software just doesn't get this. These tempo markings (by the composer; often in italian) are the track names. So Moonlight has 3 movements:

I. Adagio sostenuto

II. Allegretto

III. Presto agitato

This is perfectly highlighting that we've got no way to express our actual preferences, and are stuck treating something as a song when it isn't, but it's the only paradigm offered to us.
> Beethoven Piano Sonata #14 (“Moonlight”) is a masterpiece,

Which one ? Played by Kissin ? Or by Rubinstein ? Or by ...

Most consider the composition to be a masterpiece.

Artists/conductors 'perform' 'interpretations' of works. Whether any interpretation is 'masterful' or 'virtuosic', etc., is up to the critical listener.

Three recordings by Horowitz may have been released ... which one is 'best'? Depends on the phase of the moon and your playback equipment!

The 'quality' of a particular performance may depend on the instrument provided to the artist. The year of recording can be very important (reflects techology used). The acoustics of the recording venue are also inherently part of the performance ... e.g. too much reverb.

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.
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/

Might you release the schema into the public domain? I know that's much more easily said than done, and it's your hard work and you deserve the benefits, so no argument from me if you don't.

Or maybe share some tips on what works with the corpus of classical metadata?

While I'd love to see some of the quite spectacular internal tooling + data to be released under an open license, put to use for enhancing MusicBrainz' existing data etc., I doubt it's going to happen.

All team members have started new opportunities since quite a while and untangling all this, filtering out potentially copyrighted/licensed material etc. would take quite some effort.

I primarily make hip hop music, but I would absolutely love a platform like this where I can listen to a classical piece, and download the score and MIDI files.

Are there any legal reasons preventing this ? I don't have the ability to play classical music myself, ( my weighted 88 key sits taunting me), but I'd love to be able to take these compositions and integrate them into my own music.

IANAL, but in the US, copyright on classical pieces tends to apply to the performances, not the music. For example, a Bach piece is in the public domain (the score and probably the MIDI derivative[a]) just from being old enough, but a recording of that public domain piece may be copyrighted itself.

As always, this isn’t universal. Check your local copyright laws to be sure.

IANAL either, but. For MIDI, a public-domain work might be just mechanically converted into notes of a 'MIDI score'. Probably very flat-sounding, but technically that transcription [0] may (probably) be PD.

(E.g. My handwritten copy of a PD short story is probably PD. But my reading of it isn't.)

BUT if some famous piano player performed for the MIDI recording, then you've got tempo changes, timing, attack velocities, all that data in the file as well. Probably not PD (best to ask). (As with piano-player rolls.)

[0] https://en.wikipedia.org/wiki/Transcription_(music)

Sounds like someone needs to be awesome and release midi files of classic compositions.

You have the rare hip hop song which samples classical music directly, but I really want to learn music theory, like understanding how the notes work.

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.
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.

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.
I’d say software should handle this issue - you type in Greek and it should match all possible transliterations.
That's very hard, there are too many combinations. A better solution would be to just use the original Greek name.
That's a good solution for Greek speakers, but a terrible solution for non-Greek speakers, who _can_ learn a "canonical" transliteration but can't write Greek
There is no canonical transliteration, though. Also it doesn't make sense to learn Greek like that, and nobody does in practice.
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.
quick upvote for examples from Yanka. I thought that I'm the only one who still listening her.
>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...

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.

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.
I guess don't put on Steve Reich's "City Life" when driving in NYC.
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.
Music videos will often have different versions of the song with added sound effects, a more cinematic intro and outro, and even breaks in the song for a dramatic sequence. These flourishes make sense when you have a video but wouldn’t make any sense in purely audio context.
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

> 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.