Pandora’s Music Box


Pandora, a newly launched music service, presents what I’ve found to be the most intelligent, and accurate “relativity” engine for creating a custom online radio station.

Some time ago, I was equally positive about Yahoo! Music’s Launchcast. And I must say, I am still quite enthusiastic about it. It has widened my musical horizon significantly, and indeed, I have been introduced to a bunch of great new bands by using Launchcast.

However, the two services do differ greatly in the methods they utilize to make selections.

Of course, the exact “special sauce” is only actually known by the mutants living in the bottom of the cave, wearing tennis-ball green spandex, half-asleep in the pool of bubbling pink water. The exact algorithms are not publicly known (or at least not known to me).

But basically, here’s the deal.

Launchcast pulls music from its own database, noted at more than a million songs, and asks you to rate music as you listen. You rate artists, albums, songs, and even genres. The machine produces an increasingly intelligent mix of music as you rate more and more. If you are a serious listener and you clearly know what you like and dislike, and you go through and rate every genre, track down and rate a ton of your favorite artists, and rate religiously, you can custom create a really solid station pretty quickly.

It infers from your ratings (either 1-5 stars or 1-100) other songs you may like, by playing songs from an artist or album you have rated highly – and they play songs from similar artists, and the similarity of those artists are presumably defined by editors who manually assign the affinities.

To the extent you and another user on their network have ranked the same song highly (which happens constantly of course), another song they have ranked highly may play on your station. It’s a real-time user cross-promotion – if you both like the same song, you may also both like other songs from each other’s playlists.

Yahoo! fills in the cracks with songs simply from genres you have ranked highly (and in the event they don’t have any more music in that genre, they’ll start playing pop-schlop). It does appear they lend some weight to the real-radio commercial anthems. For example, if you rank Nine Inch Nails highly, you’re gonna hear The Hand That Feeds from with Teeth three times before you hear Terrible Lie from Pretty Hate Machine once.

Pandora, in its most essential form, is quite similar, but it requires drastically less input to get started. You give it a single artist name, then you rate songs (thumbs up or thumbs down), and can add more artists to further customize the station.

Here is the fundamental difference, and in practice, it is somewhat magical. Pandora (once named Savage Beast) uses the Music Genome Project as its algorithm (and apparently iTunes or a subset thereof as its database). It is the result of years of work by musicians, musical technologists, and aficionados, most of whom have studied music theory, most of whom are musicians themselves, and all of whom have passed a thorough examination and rigorous training in the Project ways.

This team has been cataloging more than 60 years of music across many genres – with approximately 30 minutes of analysis per song, the Genome Project records up to 400 attributes to describe a song.

The Pandora “Music Box” then takes your inputted selection, and plays songs that share similar attributes. The granularity and specificity of this level of intelligence produces an algorithm that is all but magical.

I inputted Radiohead. And this is not the place to start exploring music categorization, but let’s face it, we don’t even really know what genre Radiohead is – I mean, each of the albums is so different, there’s a couple genres on each. There are really no bands that are truly “similar” to Radiohead. I mean sure, they fit into the hierarchy of musical evolution somewhere, so there are those bands that have influenced them, there are a slew of bands influenced by them – so we’re not lost – but this is a hard one. It’s certainly a more challenging task than asking the machine to find similar artists for a standard 5-piece boom bot boomboom bot rock band.

The machine went on to play songs from many of my favorite bands, and a slew of completely new music, almost all of which I have liked. I am in shock. First, I’m critical and discerning – I’ve also been listening to music, I mean REALLY listening to it, for years. To produce so much music that I like is incredible – to dig up so much NEW music that I like is profound.

To be fair – in a couple days – maybe 10 hours of play, I have heard some dupes – so the extensiveness of the station (within the strict parameters I have so far set) is yet to be proven. Of course, by adding another 10 bands, and widening the scope, presumably that would be resolved.

They both offer free and paid versions, the free versions being supported by advertisements. However, Pandora hasn’t quite got their ad team in place it appears – I haven’t heard any ads yet. For Yahoo!, you also get a better quality stream with the paid version.

For both, you can create additional stations to play a subset of your favorite music – in Pandora, you just create a new station with new bands and groups to influence the choices, in Launchcast, you create a Mood, limited by a certain set of genres.

In both cases you can pause and skip, and in neither case can you simply select a song to play, or to go back and play a recently played song again. It’s custom radio, not a personal music library.

And listen, they are both awesome, and they both will help you discover new music.

But I rest the difference in the following example, noting “why” they have played a particular song.

Yahoo: [Song is played] “…because it matches your genre preferences.”

Pandora: [Song is played] “because it features mild rhythmic syncopation, a twelve-eight time signature, mixed acoustic and electronic instrumentation, minor key tonality and string section beds.”

Um, yeah. Exactly.

Leave a Reply