All I Have to Do Is ... Play Music
Sat, October 3, 2009 at 02:43AM
Chris Loosley in Everly Brothers, MoodAgent, Music, acoustic fingerprint, iTunes, playlist

On Music

Music, oh, how faint, how weak,
Language fades before thy spell!
Why should Feeling ever speak,
When thou canst breathe her soul so well?
Friendship's balmy words may feign,
Love's are even more false than they;
Oh! 'tis only music's strain
Can sweetly soothe, and not betray.

Thomas Moore (1779-1852), an Irish poet, singer, songwriter, and entertainer, now best remembered for the lyrics of The Minstrel Boy and the The Last Rose of Summer.

Music has always been important in my life. But lately, because we're working on Syntonetic's Moodagent launch (see the previous post), I'm spending a lot of time thinking about the relationship between music and mood.

The notion that music influences our mood is nothing new. But today's Music Psychologists are tackling the subject more systematically than the poets of old. In Seven Ways Music Influences Mood, Psyblog reviews a 2007 study of adolescents in Finland about the different ways they used music to control and improve their mood.

All the same, scientists finding connections between music and the quality of life does not affect me directly. How can I put that knowledge to use in the way I select and play my own music?

Dreamin' My Life Away

Illustration: Everly Brothers

During the week I became a teenager, The Everly Brothers recorded the song All I Have to Do Is Dream, creating a potent mix of words and music that perfectly matched my adolescent sensibilities. That song was the first record I ever bought. It burned its way into my brain, set up all kinds of connections, and never left.

Music has always affected me in that way. Once I hear a song I like, it slips into the soundtrack of my life. And like an old lover, whenever we meet again, the encounter brings back memories of the moment we first met. So it's not surprising that my tiny teenage collection of 7" singles was eventually dwarfed by a 20-year accumulation of about 800 13" LP's. Among those many new songs, some I especially enjoyed were:

Collecting Compact Disks

I still have most of those albums, lovingly preserved and carefully cataloged in alphabetical order. I even play some of my old favorites occasionally; it would take a month -- without sleep -- to play them all. But in the '80's vinyl gave way to CD's, so my collection now includes another non-stop month of music on CDs, including:

The Digital Music Dilemma

Having kept up with computer technology for 45 years, I don't like to think that I'm an old dog who can't learn new tricks; I know I can master it if I need to. So I prefer to blame my slow adoption of digital audio on my entrenched investment in yesterday's music media. But technology marches on regardless of my reluctance to adopt it, and the rapid growth of digital music downloads tells me that digital formats will eventually eclipse CDs as a distribution medium.

This first hit me when I spent a weekend in May 2006 listening to two new albums online before I could buy the CDs:

Since then I've been gradually importing more of my CD library into iTunes on my laptop, where I can play a favorite song or album whenever the mood takes me. So far so good -- but new technologies have a way of exposing problems that you didn't even realize you had.

With digital music, the dilemma is that the more albums you import into your library, the harder it is to remember all the songs you might play. Without importing any of our classical music CDs, or converting any of my vinyl LP's, I have 8500 digital audio tracks to chose from. Even after living with a distribution model based on the music album, and selecting, buying, and playing albums for over 40 years, I still can't remember every track I now own.

Playing With Playlists

My son Bryn, who's been writing and singing on his own or with Buffalo Creek since 1997, has accumulated a collection of digital music that now numbers over 20,000 songs. "I have so much music," he told me the other day, "I really don't know what I've got anymore". Lately he's been reduced to using the iTunes shuffle option to find songs that otherwise would never get played.

I say "reduced" because after you've invested years--and probably a lot of money too--amassing a collection of music you enjoy, letting a computer pick songs at random is a very poor way to select the music you're actually going to listen to. If your collection spans a wide range of genres, styles, and musicians, picking songs at random is like throwing darts at the menu in a Chinese restaurant. From time to time you may hit upon an interesting combination of dishes, but most meals will be an odd mixture that will leave you vowing to pick the ingredients by hand next time.

Smart Playlists

One solution is the "smart playlist," an iTunes feature that creates a playlist by filtering the song metadata ("tags") using any criteria you chose. If you first make sure your metadata is up to date, you can generate playlists based on any combination of rules about genre, artist, song length, your personal rating of every song in your collection on a 1-5 scale, when the song was published, when you added it to iTunes, how often it been played, when it was last played, and whether the singer was born on a Tuesday (well, maybe not the last item :). If you're inclined to adopt this approach, you can find plenty of advice online from obsessive iTunes enthusiasts.

For myself, I have two problems with all of it. First, updating song metadata is a chore. I'm just not willing to devote hundreds of hours to it. But more importantly, having studied and built expert systems for IT, I know that rule-based systems "are really only feasible (when) all knowledge in the problem area can be written in the form of if-then rules" [ai-depot.com]. You can't improve the recommendations by simply adding more rules, if the system is missing a crucial piece of knowledge it needs to make it smarter.

Picking Songs That Belong Together

When the problem area is compiling music playlists, that crucial knowledge is how the music actually sounds. Technically, this is called an aoustic fingerprint. If I can't create a rule about how the music sounds, I could spend a month adding metadata and coding a hundred rules, but those rules cannot generate suitable playlists for every occasion when I want to hear music. The rules won't relate to the music itself, to the many contexts in which I might want to hear a selection of my songs, or to my mood and preferences at that particular moment. Because I won't even know those things myself, until the situation arises.

This is why I find Syntonetic's Moodagent approach so interesting. It uses digital signal processing and AI techniques to create acoustic fingerprints that incorporate characteristics such as mood, emotion, genre, style, instrument, vocals, orchestration, production, and beat/tempo.

In my next post I will write about my early experiences with three playlist generators that aim to pick songs that belong together: Apple's Genius for iPod and iPhone, Syntonetic's Moodagent, and MusicIP Mixer, another playlist generator that creates acoustic fingerprints.

Article originally appeared on Marketing for a sustainable future (http://www.uprightmatters.com/).
See website for complete article licensing information.