My new Mini isn't here yet, so I've been driving a car w/ no Empeg for a couple weeks now... One of things I realize that I miss out on with the Empeg is a certain degree of serendipity. I choose the music that I like and listen to it constantly and often miss out on other things. I look at my TiVo, and it seems like a system to find new music might be kind of cool....

I know there are quite a few problems with this approach, but I'm tossing it out just for discussion purposes... Maybe someone has some better approaches (or maybe nobody cares about this at all ):

1) Add a rating system -3 to +3 for MP3's (this is optional, but would just make the evaluation criteria better). This could be done in jEmplode rather than on the player itself so Empeg owners don't start wrecking and dying while they try to rate music while driving.
2) Find some datasource that can fulfill the request "given that I like song X, tell me a song Y that I might also like". I was thinking I could just spider content from Amazon -- they have so much good data on user traits that I could write an app that could search for a highly rated song on Amazon, grab the other recommended albums, pick a random song off of it and save the name.
3) OK -- caveat with "piracy is terrible, etc, etc", but call this a sampling station -- buy the cd if you like it: There is a really slick java implementation of the OpenNap protocol. I could take the tune that was randomly selected and search the OpenNap network for the MP3, download it, and upload it to the Empeg.

My thought is that you would just plug your Empeg in at night and hit a "Surprise Me" button in jEmplode and it would go off and look for music for you. It just seems that if you combine the "always hear what I want" of the Empeg with "maybe there's some stuff I'm missing out on" that is targeted towards your preferences, it might make for a cool system. Basically TiVo Suggestions for MP3's.

The thing that sucks about the Amazon approach is that you could only look for songs that would be recommended based on a single other song. Ideally you would take the full set of music that you rated highly and find recommendations based on that entire set. The fix for this would be to have users put in their Amazon username/pw and use the suggestions that are already configured in Amazon for recommendations 9rather than using the mp3's on your Empeg as the original source)

Comments? Suggestions?

I've been tossing this idea around for a couple of years now, ever since I got the empeg. But I envision something that's even simpler for the end user.

Imagine the empeg keeping track of skips and replays as well as just plays. Then it would be able to infer, in a general sense, which songs you prefer in your collection. Without you having to intervene and vote, or otherwise do anything special.

Then, on the next synch with your PC, a simple connection to an online database would give you names of artists and albums which are similar to the tracks you liked, and would offer you a free sample track or two, with a "buy online" button for the rest of the album.

The only tricky part is getting the online database together and getting them to partner with SonicBlue. There are companies which are starting to head in this direction, and the Amazon system comes pretty close. A little more work and it'll be there. Amazon's system requires that you rate albums and movies before it presents you with good recommendations. The system I described above would eliminate this step by supplying a pre-created database of ratings based on the songs you skip or play.

I think the MoodLogic software is supposed to already do something along these lines, but I can't get it working properly on my system. Probably because it's just the demo version and doesn't really do anything.

This system would be the first step towards my idea of a utopian future where all artists can connect directly with their fans, regardless of how obscure or non-mainstream their music is. Because this online database would be looking up music based on what you really like to hear, as opposed to what the record companies think you should like to hear.

That sounds like an absolutely great idea. I use the amazon recommendations all the time to find other music that I might be interested in and it seems to work pretty well. A jemplode maintained playlist of "songs I might like" would be an amazing feature.

-Mike

I am surprised nobody else has chimed in on this thread yet, it really seems like a neat idea.

I was thinking about ways to get a larger datasource than the single song method you mentioned. How about scraping the "related artists" section of AllMusic or a similar site? Make a list of all the artists on a users empeg (or somehow select a subset of the artists), query amg for all the related artists (this would take a while). Build a rating system based on which related artists where returned the most times, and if the user does not already have music by that artist download some.

I guess this is similar to what you already outlined, but at the artist level instead of the track level. Plus this gives you the ranking across multiple samples instead of a single sample.

-Mike

I'm also surprised there hasn't been more response to that post ... But I guess I don't care, since I'm going to write it anyway

I like your idea better than mine ... I've never seen AllMusic before. I was also thinking about maybe getting better track data from MusicBrainz, but i suppose just having artist data would be enough to search for tunes.

ms

But I guess I don't care, since I'm going to write it anyway Sweet! As always, let me know if you need a tester once you get something going. -Mike

Imagine the empeg keeping track of skips and replays as well as just plays. Then it would be able to infer, in a general sense, which songs you prefer in your collection. Without you having to intervene and vote, or otherwise do anything special.


The mp3 streamer I hacked up for use at my house long before I had or heard of an empeg did just this. I never considered extending it for "find music like this" but instead it was intended to provide a weighted random playlist system, where based on "who" was listening and what we knew of their listening patterns, we'd weight the random selection of tracks according to their preferences. So, for instance, the system would be less like to play any Alanis while I was listening than the listening preferences of say my wife would otherwise cause it to.

Like so many worthwhile projects it died due to "no time".

I definately agree with you on the no time issue. I've had loads of projects which have died a slow lingering death because I've just never had the necessary time to finish them. I also did a MP3 streamer prog which used IceCast to send the data. I think I got it to stream and to load playlists and to control it with a web interface before abandoning it.

- Trevor

Yeah .. I commented on my "graveyard project pile" in another thread. There are so many ideas and so little time. When I am working towards a release, jEmplode alone almost becomes a additional full-time job. I have so many half-started projects that I'd love to work on ... That's why I'd love to gineau pig that drug Provigil (basically makes it so you don't feel like sleeping for days at a time without the side effects of, say, Cocaine ).

Mike

Mine used the xmms perl modules to control xmms, with the liveice xmms plugin sending the stream. Why? Well, would you believe the crossfader plus the volnorm module made xmms the best choice for a streamer at the time? Heck, it might still. And the web interface for the request system was hacked up in perl with the script backends reading a postgres database, that also being populated by perl scripts. At some point I'd like to make the web frontend to the database able to upload files to the empeg, but, what time??

If the scripts would be amusing to anyone, I can put them online again somewhere.

I killed all my free time simply; In addition to my "real job" I first started volunteering on a distributed filesystem project (OpenAFS) then got a second job on the side providing some AFS development work, which basically means I'm hosed. Sooner or later I'll have to decide extra toy budget (since I'm not managing to actually pay off any debt ) isn't worth it and stop the insanity, and go back to only volunteering. Maybe you don't get paid but the time demands aren't quite as demanding.

On the other hand I keep debating toying with how I can do an in-kernel AFS server for the empeg. It's not really worth it aside from hack value.