Beyond Sound and Science: Musicians, Researchers, and the Next Spotify

Beyond Sound and Science: Musicians, Researchers, and the Next Spotify

Musicians and researchers must work together to build a streaming music service that is musically and technologically sophisticated, appealing to audiences, fair to musicians, and conducive to direct engagement between the two. We must be involved in this next phase of creation, or we leave the future of music up to others.

Written By

Isaac Schankler

Last week I was filled with excitement and trepidation for the future of music. The Computer Music Modeling and Retrieval (CMMR) conference was held at Queen Mary, University of London, bringing together “researchers, educators, librarians, composers, performers, software developers, members of industry, and others with an interest in computer music modeling, retrieval, analysis, and synthesis.” If you’re not familiar with the field of music information retrieval (or MIR, or Music-IR), much of its current focus is on the automatic extraction of musical information from audio. Perhaps the most obviously apparent use of MIR is Shazam, the app that recognizes what song you’re listening to.

Musicians need to be paying very, very close attention to what researchers in MIR are doing right now, because it will have a dramatic impact on how music is created and consumed in the near future (even more than it already has). CMMR’s Thursday panel discussion (with John Sloboda, Alan Marsden, Tim Crawford, and Joydeep Bhattacharya, moderated by Geraint Wiggins) was particularly illuminating in this regard. Somewhat unusually, the larger impact of MIR was queried and prodded—Crawford talked about how the field isn’t serving musicologists, Sloboda discussed how it’s not serving society at large, and Marsden addressed whether or not it’s respecting the musical artifact itself. These are all important and provocative points, but almost shockingly, no one mentioned the impact of MIR on musicians or composers.

This omission is even more curious and troubling in the context of the recent heated debate centered around David Lowery’s Letter to Emily White, an alternately heartbreaking and exasperating article about the role of technology companies as the new arbiters of musical consumption. Unfortunately, some of Lowery’s best points are obscured by his patronizing tone, and his moralistic scolding of younger generations for their listening and purchasing habits seems unlikely to change anything for the better.

However, there is no doubt that Spotify, the target of much of Lowery’s ire, is bad for musicians; the only real question at this point is just how bad. The reasons for this are plentiful, but most of it stems from the fact that services like Spotify disconnect musicians from listeners. The more anonymous music is, the less likely people will be to feel attached it and to feel the need to support it. But when someone knows who you are, when you’re not just some disembodied vibrations in the air, they’re far more likely to stand behind you. It’s no coincidence that the musicians who have had the most success on the internet are those like Amanda Palmer, who aims for a very direct, personal connection with her audience. And it’s important to note that technology doesn’t necessarily lead to alienation; Palmer has leveraged services like Bandcamp and Kickstarter very effectively to create and maintain that sense of connection.

And it’s not just the warm fuzzy feelings either. There are very practical reasons for preserving the direct relationship between artist and audience—when you put someone in the middle of that exchange, you cannot necessarily trust them to have your interests at heart. With Spotify in particular, there is not one middleman but two: the label or distributor, and the technology company. It’s no wonder that the artist is the loser in this transaction. Charles Caldas of Merlin, who negotiates with Spotify on behalf of independent labels, has released some growth percentages for payouts to artists, but conspicuously, no dollar amounts. Zoë Keating, on the other hand, has been remarkably transparent about how she’s been compensated by various streaming services. With over 40,000 plays, Keating earned a seriously underwhelming $150 from Spotify for an entire quarter (12/29/11-3/28/12). In terms of popularity, Keating is pretty much at the top of the heap as far as independent classical-ish performers go—if you make music that is even remotely similar, you are extremely unlikely to do any better.

But is this really any better than other internet music sales models? Well, I’m not nearly as popular as Zoë, but enough people have listened to and bought my music on Bandcamp to make some sloppy calculations. Personally, I’ve made about $0.02 per play, which is over five times as much as what Spotify doles out. This isn’t a perfect comparison, because Bandcamp doesn’t pay per play; someone has to explicitly decide it’s worth it to pony up for my music. But even if only a small percentage make that choice, this model is still vastly better for the artist financially. (I’d be willing to bet that Zoë Keating’s play/buy ratio is better than mine, too.)

Supporters of Spotify are quick to point out that musicians get publicity from others sharing their music, but currently, there’s no real way to measure the impact of Spotify on how people share music, so this all remains conveniently speculative. And if Spotify becomes the main outlet through which people listen to music, this won’t matter much anyway.

Which brings me to my next point: Musicians need data about their listeners. I want to italicize and bold and ALL CAPS this point to the heavens, because I think it’s sorely overlooked. All the back and forth arguments about whether or not streaming music is good for artists is completely hypothetical until we have hard data, and most companies are loath to give away this proprietary information. The lag between when things happen and when artists find out about them is another huge problem—what’s the use, really, in finding out what music people were listening to over a month and a half ago? It’s hard to act on that kind of data unless it’s recent. (Bandcamp and Soundcloud are exceptions in this category, giving more or less immediate feedback on what people are listening to. Bandcamp also tells you how long they listened for, and what site they came from.)

This is why musicians desperately need to have MIR researchers on their side, because if there’s one thing they know, it’s data: how to get it, what it means, and how to use it. They will also almost certainly have a hand in whatever the next big music streaming service is; the theme of this year’s CMMR conference was “Music and Emotions,” and one recurring idea was a Spotify-like database that selects music based on mood. Fortunately, researchers also desperately need the expertise of musicians, though many are just beginning to recognize this fact. One of the most commonly evaluated MIR tasks is automatic audio-based genre classification, and while various state-of-the-art methods have been thrown at this problem, the accuracy has hovered around 60-65% for the past four years, an epoch in research community terms. As a result, researchers are finally acknowledging that methods based on surface musical features may not be enough, and higher-order musical knowledge may actually be necessary.

For these reasons, as much as I’m sympathetic to Lowery’s position, I can’t agree with his analysis of the current situation. I don’t think that the solution to today’s problems will be found in the consciences of individuals, which are so easy compromised by convenience and circumstance. I think that the solution needs to be one of infrastructure—musicians and researchers must work together to build a streaming music service that is musically and technologically sophisticated, appealing to audiences, fair to musicians, and conducive to direct engagement between the two. We must be involved in this next phase of creation, or we leave the future of music up to others.