Academic Productivity

How can a music playing program be a time saver? What does this have to do with productivity? Well, background music prevents me from getting bored and drift into distractions. Music may shield you from noises and attention-grabbing events around you. I think music helps me reaching flow when writing/programming.

I will assume that at some point you have taken the time to rip your music collection into your HD, and that you have decent tags. Changing CDs or vinyl is just too distracting. If your tags are a mess, there are lots of tutorials on the web to get them under control. It’ll be worth the effort. At the end of this post, I’ll show you what could be the fastest method with the least human intervention.

(Note: I have talked about how managing music and academic paper collections are similar here; See also ‘noise for academics‘ by Shane).

The problem is that having background music has a cost.

The cost is that you have to keep music going, and DJ for yourself. For a start, you need to make some decisions as to what music is appropriate (no lyrics, no high-novelty, not too emotional -kind of predictable- works for me). And this means that sometimes you need to stop what you are doing and go change the music. With computers, this is less of an issue, as programs can keep a playlist. Never worry about having to switch CDs, or having a ‘sticking out like sore thumb’ song in a CD that would take away your attention. But still, making the right playlist takes time (and you want to explore your collection, of course, not playing the same thing over and over again!).

Another attention-grabbing distraction is a sudden change in volume; I recommend to use replaygain in your collection to prevent this. Replaygain is available in most players, although most use implementations that are non-standard. The closest to a standard is what Foobar provides. You can replaygain your collection in batches. It’s CPU intensive but you only have to do it once.

The old-style approach of dragging-and-dropping a buch of hand-picked songs into a playlist by hand is time consuming, does not encourage music discovery (you tend to put in things that you already know! otherwise, you may hit a lot of distractions in the new ones!) and it

Now, we have a well-defined problem: we need to create playlists suitable as working music. We have a definition of what’s good for work: no lyrics, no high-novelty, not too emotional -kind of predictable- (your definition may vary). And we want to generate new playlists, not to recycle the same one over and over again!

I think solving this automatic DJ problem can save me a few minutes a day, plus it can make me enjoy my music collection more, so it’s a problem worth pursuing.

At the beginning of my search for a solution, I found three main recommendation services, with very different methods: Last.fm, Pandora and MusicIP. They represent different representations of music knowledge: usage patterns, hand-coded features (supposedly assigned by experts), and acoustic analysis.

Last.fm

Last.fm is a tool that tracks what you play in your computer; then, using your profile you can find your musical neighbors and look at what they are playing. Since half of the world with an Internet connection has been tracking their music playing with last.fm for years now, they have a succulent dataset on which one could build a nice recommendation engine. But when it comes to pragmatics, this approach seems to be flawed. Here is how.

Last.fm has an amazon-like feature “users who liked artist XX also liked…”, and a list of users that are the closest to you in musical taste. Great. But, while looking at your neighbours is useful and fun, it involves a lot of footwork. Ideally, I’d like  this to be automatic, so I can push a button and start writing. Looking at your neighbors is ideal as a way of expanding your music horizon, but it is a source of distraction, which is exactly what we are trying to avoid. There’s an additional issue: you may find new music, but this is not in your collection yet! You need to acquire the music, and additional step and a distraction in itself.

Last.fm also provides you with a ‘personal radio station’. It can stream music from your collection, using parameters such as how often you played it and when was the last time you did. While this is interesting, it will use music from your collection that you played, AND music similar to that in your collection -recommendations- BUT without knowing if it’s in your collection or not! I want to explore the music I already have, not be enticed to get more (the average user plays only 15% of their music collection over and over again; a waste of resources if you ask me).

Pandora

Pandora is a recommendation service that builds a personal radio station for you. You have to seed it with a band/song that you like (note: Pandora systematically ignores classical!). Then, it’ll bombard you with music that it thinks must be similar to your seed. You can input “I like it” or “skip it”, and this way recommendations get better with time.

The key question is how Pandora knows what is similar to what. Well, they have used a somewhat ‘brute force’ approach: paying humans to tag every piece of music known to man in some scales -proprietary format-. This is called the Music Genome Project. There is a list of Music Genome Project attributes for the curious. Again, Pandora is not that good to explore music you already have, nor to create playlists that you can ‘set and forget’: you are always at risk of hitting a super-distracting song. You need to be online to use Pandora, which is not compatible -in my view- with being in ‘writing mode’.

On top of that, last time I tried to use Pandora I was greeted with a message saying that under the EU laws, they could not provide their service anymore. So, not much to see here anymore.

MusicIP

MusicIP does many things right, and it could easily be the next revolution in music players. Forget iTunes, WMP, winamp, and even foobar; this thing used as a player can give you functionality that would literally be ‘in your dreams’. And the main focus of the company is not even making a great player (!). They key contribution of musicIP is their music fingerprinting/audio analysis. Again, this is proprietary technology; I’d be very curious to see how this works.

MusicIP will scan your music collection and generate IDs for all your songs. This is a CPU-intensive  process that can take several days. As soon as you point MusicIP to your music folder, you will see how most of your titles turn green. That means that, without doing the analysis, MusicIP found a match in their database. Any green song is immediately mixable. The entire process does three passes through your data. Quoting W. Hicken, main developer:

The fast initial pass is based on metadata – we have mountains of metadata correlated with acoustics, so we can usually get pretty close based on just that.  The slower pass is to verify the metadata isn’t lying (so if you’ve got a song which is completely mistagged, the acoustics originally assigned will be wrong, but fixed in the slower pass).

Do you like the song that you are listening to? Do you want to see what you have in your library that’s similar? Hit the mix button on a green song. That’s it. A new playlist will be created… and surprisingly, it’s pretty good. More than enough to just leave it playing and get to work.

But having your entire collection fingerprinted has many advantages. You can use these IDs to do advanced searches. Since music representation is not limited to artist/album etc tags, one can actually think in terms of ‘semantics’. The search features are incredible. You can create custom tags. You may think that this is not too exciting.

have a look at how you can define custom searches here.

Note that fixing missing metadata is a time-consuming task… well, the beauty of having a musicIP-fied collection is that you can use their gigantic database to fix your tags! Just select a block of songs, and hit the ‘fix tags’ button. Since they collect tags from many user, the correctness of them all is powered by democracy

Now, to keep the writing flow, I need an application that lets me skip songs, change volume, etc with shortcuts, without having to chase the application window. In MusicIP you can assign keyboard shortcuts, and make it very unobtrusive.

Power features

I forgot to mention that you get all those features for free!. For those considering going for the registered version, jump in!
The new power search features, moods, etc are really cool. The best way to learn about those new features is the musicIP blog and some user blogs.

Basically, you can create a playlist that you really like, save it as binary mood, and send it to a friend. Even if he doesn’t have the same songs, this binary mood will select similar ones from his library.

You can generate as many playlists from your library as you want using your mood settings. This is the nirvana I was looking for to create ‘a soundtrack for writing’ automatically.

There are many more features, but it seems that musicIP is one of the most under-documented applications out there. No wiki, no comprehensive help (just pages scattered over the site talking about different topics), some blog posts, and that’s it.

Cons: MusicIP can see your replaygain tags and use them if you have generated them using foobar, but it cannot create them; you’ll need another music program to scan you collection (a one-time thing) and apply replaygain.

This entry was posted on Sunday, September 2nd, 2007 at 5:07 pm and is filed under Blog, Evaluation, Socializing, Software, Writing. You can follow any responses to this entry through the feed. You can leave a response, or trackback from your own site.