(That’s right. I’m get paid by the alliteration, bitches.)
First some context: this weekend DKraft and I were bumming around Boston, shaking off hangovers and pre-gaming for a local show that night, and we logged onto the IntraWebs and messed around with Pandora for a few hours. For those of you un-familiar with Pandora, it’s basically a website that streams a song or band you type in and then plays continuous music that has similar characteristics (i.e. a similar sound) as the original band. Based off of the well-known Music Genome Project, it’s a pretty ambitious project to break down individual songs or bands into tangible, or at the very least, describable characteristics and cross-refererence them with other sounds. It’s designed for indie rockers like us here at MaxEnt to be able to discover new music with similar styles as the music we already like, for those people who crave new, good, interesting, unique music. So DKraft and I typed in some of our favorite bands…Mountain Goats, Built to Spill, The Hold Steady….and waited patiently for…
…nothing. Surprisingly, nothing. Sure, there were a few songs that were tolerable, even stuff that I would give a second and third listen. But for the most part I was really struck by how little the songs that Pandora chose for us were anything like the original song/band. I really liked the concept of the project, and was (and still am) willing to keep trying it out to check out some new stuff. But it did get me thinking. What makes the music I listen to, well, the music I want to listen to? What makes Neutral Milk Hotel give me such a strong, almost visceral reaction everytime I listen to them? Why don’t I get that same reaction to the twenty-some other bands that either list NMH as an influence, or vice versa? Could I describe 17 different distinct, autonomous qualities that makes this reaction happen? Or is it just a random combination of external forces (i.e. a soundtrack to good times with friends, relevant lyrics to what’s happening in my life, etc)?
Ok, so most of us have thought this before. And I could easily extend this to a prodigiously pretentious discussion of Pandora and the futility of objectively catagorizing art. But that’s not what I want to do. And part of me kind of thinks that it is possible to empirically describe art. That’s right. I said it. But Pandora just wasn’t quite making the connection for me. What were the qualities that they were using to characterize songs? How were they gathering data about the bands/songs? How was that data translated into easily processed and manipulated raw numbers that could be worked into a usable interface? And how the hell was I supposed to find the next Hold Steady when all I kept getting was fuckin’ Superchunk and Cheap Trick?
Fast-forward to this week. I’ve been working a client of mine for a while now, and recently we’ve been talking about his work on his new venture capital project. He is not one of my typical clients. That is, he finished high school. Actually, not only did he finish high school, and college (and two or three Master’s degrees), but he’s a former wunderkind start-up genius who had the world at his feet. (Suffice to say, it’s been downhill since then, which is why he is my client.) His project that he’s been working on is a self-proclaimed, bigger, better, and more advanced version of Pandora. And in the past few days he’s started to show it to me. I told him I was a big music fan and played some music myself. So he’s been using me as tester for his new software, software that is absolutely brilliant and frighteningly accurate. (Again, because of confidentiality I can’t reveal his name or his company’s name. Yet.) I don’t understand a fucking thing about algorithms, programming, or software engineering. All’s I know is that, despite some pretty serious bugs in the programming, this guy is on to something huge.
He used to know some of the Pandora guys and he says they are doing it all wrong. He says rather than comparing what two musical bands or songs have in common, we should be looking at what they don’t have in common. Take Dashboard and Mountain Goats. For me, I can’t the thought of listening to one, and the other is one of the more unique bands out there that I listen to almost on a daily basis. He sat me down and showed me how and why Pandora would show these two seemingly disparate bands actually have quite a bit in common. Male, singer/songwriters, acoustic guitar-strummers, focus on melodic structure, minor key hooks, etc. The list when on and on. He’s been talking about how the algorithms are structured, and by no means is it perfect, but it is a glimpse to how to objectify the music that each of enjoys.
So what do you use to differentiate between bands, anyway? Some pretty interesting qualities, actually. I can’t go into details (patent pending and all) but there are some algorithms about vocal tonality and how it compares to the “Clear-Channel-defined quality of what is “marketable” and what isn’t”; a more expansive examination of lyrics, lyrical content, rhyming patterns, and modes of delivering lyrics; and most importantly, a way for listeners and musicians alike to describe their music with a series of hundreds of descriptive words/feelings/moods/etc.
So what does this mean? This product is still in its infancy, to be sure. But I know that it starts to raise some questions about how we are begining to be able to pinpoint, in somewhat of an objective manner, what makes music “good” for each individual person. It is not a fool-proof project. Inherent in art, especially music, is the factor of individual preference for some un-identifiable reason. Ultimately, what gets me to turn up the speakers, down my drink, and dance around like a fool is something that might be nigh impossible to exactly pinpoint.In addition, it becomes even more of a challenge when it comes to very “unique” bands that don’t necessarily wear any of their influences on their sleeves. For the bands that have carved out a unique and distinct sound, how can we describe it without relating it to other bands? But the fact that there is so much music out there, I think that software that can take 37,000 bands down to a few dozen for someone to make a decision about what they like is a step in the right direction.
I’ll continue to update this section as I learn more (confidentiality permitting, of course). Until then…