Why Can't Streaming Services Get Classical Music Right?

What makes listening to anything other than entire albums (radio, streaming, etc) unpleasant for me is the volume settings.

Most people listen to classical way too quietly. Most classical music has a very wide dynamic range. Your volume should be adjusted so that the very quietest parts are loud enough to hear every subtlety and so that the loudest parts shake your bones.

A single trumpet playing at full volume is loud enough to cause hearing damage. An orchestra has several of them, along with 50+ other instruments.

Loud orchestras are loud enough that freaking cannons do not necessarily overpower them.

So next time you're listening to classical music (or jazz), crank it up, and discover how much more you enjoy it.

Unfortunately, you'll ruin streaming and radio, though. They're often compressed, ruining that huge dynamic range. Each piece will have a different level, so you'll constantly be playing with your volume knob to get it set correctly. Which is really hard if you've never heard the piece before -- is it supposed to be loud or quiet?

And if you're listening to the radio with the volume set correctly, the announcer will start screaming at you. I love the CBC's Julie Nesrallah's voice, but I can't listen to her program because they have her voice turned up way too loud. If they turned it down the program would be much more enjoyable, and people would enjoy the music more because they'd turn their radio up to hear her properly.

They're simply not built for it. Most streaming music services are built for people that want to listen to deadmau5, g-funk or top 40 while they put on their party dresses and pre-drink. A lot of us also use them to stream random tracks while we work: stuff with no lyrics, etc.

But a bespoke solution for streaming classical music should ideally have metadata fields for all those things: conductor, soloists, orchestra, recording venue, date, etc. etc.

I'd be willing to bet that internally most of those services are using nothing more sophisticated than the id3 tags I remember mucking with in winamp back when Napster was still a thing.

This is actually only half the problem. Classical music is one of the genres where the profit-sharing model doesn't work very well. Spotify pay out every time a track is listened to (IIRC, this is triggered when you're 30s into the track). This works well for pop music, where the songs are short and they are listened to many times by the typical user. It works less well when the "songs" are 45 minutes long and typically only listened to once (famous works are recorded by multiple orchestras and listeners are likely to check out several and perhaps only ever return to their favorite interpretation).

This means that for many it's still more profitable to sell a small amount of CDs than to be on spotify.

I suspect this will change over the years because there's now a very large overlap between those that still listen to CDs, don't use spotify, and listen to classical music.

Bad metadata is the plague of the current long-tail world of Internet commerce and services. Classical music is a fantastic example, as is map data, online shopping, and pretty much everything else on the Internet.

The fact is that good metadata requires human verification and editing, and editors are the last thing any Internet company wants to spend money on.

Strangely, I find that most of my classical music is streamed acceptably (not perfectly, but good enough).

Try listening to Abbey Road, or Dark Side Of The Moon, or any other album that relies on zero second time gaps between tracks with a constant beat (The most recent example I can remember would be Chemical Brothers, "Further", between track 1 and 2). That is annoying, and has been throughout the history of stream-able audio.

Why can't we get canonical track timing for albums (down to the sample) and base the playlists accurately off them?

Does anyone else have this problem, or am I doing something terribly wrong? I've come across this problem on iTunes, WMP, Spotify, VLC, Google Music, Soundcloud and pretty much any other digital audio player out there that isn't designed for production (Traktor, Serato, DAWs). The fact that Traktor and Serato get this perfectly right hints that the "consumer" products just don't really care about this problem.

This reads like a spec for a startup opportunity. It's so clear to any one of us what the data model needs to look like and how to make search work. The only real issue is the human labor needed to do the data entry, and the licensing.

The market is huge though. Whoever steps up and attacks this with VC will do quite nicely.

I wouldn't be surprised to see a startup launched from this article in the next YC round.

Similar thing for Japanese music.

Should I search for 椎名林檎, the Japanese spelling, Shiina Ringo, the transliteration following standard transliteration rules, or Sheena Ringo, the transliteration that the artist has chosen herself? (It's the latter, which was unknown to me)

Moreover, you might be searching for the phonetic spelling using Japanese kana, しいなりんご. That's a lot for a system to handle. Maybe Chinese people will search for her using simplified characters (not for this one in particular, but definitely others)

Even native music players don't get classical right. It's virtually impossible to shuffle through a classical music library because you always get plopped in some movement in the middle of a piece. I've always wished that there were a way to group certain pieces together so that all the movements are treated as one piece for the purposes of shuffle. I remember someone else making this complaint about three years ago and, to my knowledge, nothing has happened since:

http://sigma-star.com/blog/post/startup-idea--classical-musi...

I've been frustrated about this for quite a while. About a year ago, I decided that I should just try to roll my own (just for me, using my music collection).

I've since open sourced my effort here:

https://github.com/tchaik/tchaik

What it does (so far):

- prefix grouping (so that Symphonies/Concertos etc are automatically grouped together, and the groupings transfer into playlists etc - so you can queue a whole symphony at once). - enumeration detection. - splitting multiple artists into separate fields (common where you have Conductors and Orchestras as the artist). - reads iTunes XML Music Library files, or will extract metadata from supported audio files in a directory tree (see https://github.com/dhowden/tag for supported formats!). - store your music in the cloud (Amazon S3 supported), locally (on the same machine that hosts the UI server), or on a remote file store. - web UI (using ReactJS), music played through HTML5 audio.

Amongst the things that are still in the "plan": - gapless playback (HTML5 makes this a bit tricky/messy). - Opus codec support (for streaming to mobile devices). - many more things!

"The other huge issue, in terms of classical streaming, is sound quality. It stands to reason that picky, "elitist" classical music fans would also be picky about audio standards as well. And while it's entirely true that bit rates don't matter one whit when you're listening through standard-issue earbuds, most of the best-established current services don't emphasize great audio quality. Mahler's epic, sweeping Fifth Symphony, for example, is a watery shadow of itself when I hear it (listening on very good headphones) at 160 kpbs on Spotify's free service. Lossless sound is one of the biggest points of differentiation that Tidal is trying to make for itself, but so far the scope of their classical offerings and the quality of their metadata have been a disappointment."

This person has never ABX tested 160kbps lossy versus lossless. This bogus and unfounded claim calls the rest of his expert assertions into question.

Even on ideal headphones/amp, I know no one who can consistently identify 160kbps from original PCM in an ABX test.

She has a point, but overall I'm reminded of Louis C.K.'s "Everything's amazing and nobody's happy" schtick. I've gotten pretty deep into classical music - consuming a dozen or so books on the subject and listening to nearly all of the referenced works - essentially because of streaming services. I would have had to spend thousands of dollars to be able to listen to all the classical music I've listened to in the last 2 - 3 years.

Yes, finding a specific recording is sometimes hit-or-miss, but I've found world class recordings of just about everything I've looked up, and it's almost always easy. It's pretty easy to spot a Deutsche Grammaphone or Naxos recording, and I find that the services are getting better and better at surfacing high quality recordings when you search for a given work.

It's odd that she doesn't mention Apple Music since it has been in the news - I find their metadata and presentation of that metadata to be worlds better than Spotify or Rdio.

It's quite ranty, seems the author didn't really do much research - in this thread there are several promising classical only streaming services mentioned -

http://www.contraclassics.com http://www.idagio.com http://composed.com

Composed look like they will be doing a good job - https://composed.recruiterbox.com/jobs/fk0hpm5/

The recommendation engines on Spotify, Tidal etc are usually rubbish for anything, so I'm not surprised it's the same for classical.

I've found Youtube to be the best place for more niche music, and their recommendation engine is pretty good. Sure it's not super high bitrate but does the job for me. Ads are the only annoyance.

The metadata is provided by the labels. For their top-selling popular music that they earn songwriting and performance royalties on, the metadata is typically very thoroughly checked and accurate (although there are still mistakes). For classical music that is long out of copyright for the music, they care a lot less and the metadata is filled with mistakes. Most streaming services make a decent effort to try to clean up the metadata, but to a large extent, it's just a case of garbage in, garbage out.

I had exactly this prblem when I signed up to Qobuz yesterday. Just as the author did, I typed in 'Mozart magic Flute' and got all kind of crap back in no specific order. They did however allow me to select a specific CD and sort of 'save' that.

'Brahms Clarinet' didnt do well at all.

Metatdata designed for pop / rock doesn't work for classical.

I cancelled my Qobuz sub after a few hours on the basis that they didn't even TRY to solve this problem.

WCPE in Wake Forest NC streams classical in Windows Media, MP3 and Ogg Vorbis formats

http://theclassicalstation.org/listen.shtml

As classical fans themselves, they pay more attention to artists, composers, and performances. But there's still a limited amount of info they can announce.

The author goes on a lot about metadata and bad search interfaces for it. Yes, classical music metadata doesn't fit with the "simple" pop music artist/album/song format. (Though the "featuring" and remixes in pop music also mess that up.)

Often people try too hard to fit data into a fixed schema or ontology. It made sense when data storage and retrieval was based around tabular and relational data, but in these days of search engines and document databases and JSON, it's just unnecessary. Instead of pre-determining the fields you search on, you should search with free text, then refine with whatever metadata happens to be available.

So instead of messing with Spotify's "artist" search, you could just go to Amazon (or Google) and type in "Beethoven's Ninth Bernstein New York Philharmonic" and you'll almost certainly get back the items you want near the top of the list.

i think alot of the suggesting issues is with the collaborative filtering model often used in suggesting songs in a streaming music service. The collaborative filtering model is not well suited for communities with extremely high variance within a specific genre. In other words, in the context of the model, classical, and classical sub genres are not differentiated enough, in terms of individual interest, to form a new independent cluster. The example being, I like romantic era classical, but i don't thumbs down at the first note of a classical/period classical piece. It's that openness that retrospectively is killing the model. A human curated classical playlists like on 8track, I suspect produces a better experience (of course depending on who put the list together). Crowd sourcing beats collaborative filtering here it seems.

The very same issues plague modern music too. I listen to a lot of electronic music, which comes with a lot of people making remixes and such. Those are already a massive pain.

Then i also listen to vocaloid music, where the complexity gets downright insane. Vocaloid music is music where the "performer" is a software using a prerecorded soundbank to render the vocals into the final piece.

For those you typically have: The song writer, the music writer, the music arranger, the vocals arranger and one to many soundbanks (which may come in variants, with two soundbanks being recorded from the same person, but with different tonal qualities). And to put the cherry on top, most of the people making this music only put them on youtube or niconico, the japanese youtube.

Getting sound files with any sort of useful metadata for those is almost impossible.

There is http://www.contraclassics.com. It's a Norwegian company and the service is designed and developed by http://nordaaker.com

Urgh, I made the mistake of trying out the Classics Online HD.LL mentioned in the article.

I had to make an account before I could actually find out the payment/subscription details, and only after getting that far did I discover that the "web-player" requires a special streaming plugin with no Linux port.

None of this was helped by the website being godawfully slow to do anything, and it popping up a box asking me to install the plugin every 10 seconds when I wasn't even trying to play music.

Finally I get sick of all of this and try to delete my account, which is, as far as I can tell, impossible.

I actually have this exact same issue with managing my music collection. It's all about getting the metadata consistent and there's no easy solution for that.

This sounds like a problem that could be easily solved through crowd sourcing.

1. Set up your metadata format with enough fields and flexibility.

2. Let your users help you fix your metadata, wiki style.

3. Profit

I have heard advertising for composed.com, which specializes in streaming classical music. I have not tried if they get it right.

Is classical music distribution a problem that streaming can easily solve? In pop music, you have people singing in different national languages in perhaps hundreds of commercially viable genres. Classical music certainly has genres (opera, early music, modern), but it's a coherent tradition, so switching between subgenres is more like switching between classic rock and contemporary rock than between rap and country. Moreover, classical music is in wide supply, as many mid-sized American cities have a quality classical radio channel (say, on NPR), regular amateur performances, and even professional orchestras and other companies. Besides, good-quality classical recordings are cheap and widely available in a wide variety of formats.

EDIT: My point was that perhaps the benefits of a good streaming service are lower to classical consumers (than pop music, which focuses more on unique individual performances). If the costs are higher and the benefits lower, that could be a sufficient explanation.

The internet makes the technical problems of the long tail almost vanish, but maintaining all that metadata is still work. And the music streaming services make enough money in the mainstream with practically no work at all, so why should they care?

If I'm not mistaken Classical Archives does streaming:

http://www.classicalarchives.com/

Pareto Principle?