Th present project on my mind is a “Web Application” that simply ‘recognizes’ your ‘taste of music’!

You’re interfaced about 20 “always random” songs from a database; you play them as you like; and after collection of enough info, the web application, using AJAX (so that your play of songs dosn’t gets disturbed), suggests you a few more songs based on the taste you’ve been adhering to.

Any more addon ideas?

Abhishek
(India)

Sounds a lot like the Pandora Radio service.

Still in all, I think it’s worth researching. I’d be interested in seeing updates on your project.

I have a similar project in the works using Slide Agent but with books instead of music. You sample content from books of your interest and more books are suggested based on your rating. Dave is right about your project sounding a lot like Pandora and several other virtual playlist services that are out there.

Sounds like a fun project though.

You’re going to need something like this.
Don’t know if it’s a good idea to do this in javascript speed wise unless you do the calculations on the server, off-line.

Well, the Pandora Radio service seems to be restricted in India.

And, Jan, I just opened the link, and it seems to be a high-level article. I don’t even know if it’s of my level. So, should I be going for it (as what all I saw in the article didn’t seem to go inside my head!), or can you tell in short, what it is all about; I didn’t understand.

And, yes, I’m doing the work server side. It’s based on PHP.

Presently, as a basic structure, what it does is that, it refers to a database of songs, and let’s you listen to them. And, it actually tracks the “type” (eh, naively) of music you listen to, and then, again queries the database, and selects songs of the very same type, and provides them to you as a suggestion through AJAX.
[The songs are simply played on the application using HTML5 Audio API].

So, how’s it, yet?

I thought you were going for a more advance approach and analyse songs to look for similarities. The article is about how you can find duplicate songs, which is a more stricter form of ‘similar’. hence the link.  Doing this type of audio analysis requires lots of FFT calculations or worse still by using a series of low-pass filters. This type of processor intensive computation is best done with MMX, SSE3 instructions or on the GPU, which I don’t think PHP or javascript can do.

That said, if you’re going for a simple ‘music-style’ db, you don’t need all of the above.

Yeah, I understand that. But, what more should I be doing then, as my case is with PHP, and might be a required amount of JavaScript?

Also, I desperately want to give in it at least some sort and amount of Artificial Intelligence.

What can I do for that? I’ve been always desiring for this.

So you’ve heard of Amazon.com and iTunes, right?

Yes, why?

Because they make recommendations based on your previous habits.

Yeah, may be, haven’t used them though.

But, I want some suggestions to put in some actual “artificial intelligence”. What should I think of, and start working on?

Abhishek, if you don’t want to get into the type of spectral analysis Jan mentioned, I suggest the following: come up with a series of descriptors with which to rank songs. Perhaps by vocal style, rhythm, instrumentation, etc. I’m sure there are typical categories in music theory that you can research, or just make them up yourself.

Categorize a set of songs using these descriptors as a basis. Each song will have a “vector” in this basis (basically a list of values, one for each descriptor). Let users rank how similar two songs are, with one being from your set, and come up with a scheme to adjust the basis vector of each new song according to how similar it is to various other songs in your set. After you have a library of music with assigned vectors, you can use the common vectors that appear in a user’s music selections to recommend new music with similar vectors.

The scheme with which your system updates vectors according to the user-provided similarity score would be the artificial intelligence component. Some sort of Bayesian approach would work well, I imagine. A similar system can be used to refine guesses for new music based on the vectors in the user’s music selections.

By the way, the reason you are getting terse answers is because there are already (very good) systems out there that make selections based on user preferences. Many have been mentioned in this thread. A very good start would be for you to take some initiative and search for information about how these systems work!

there are so many algoritms around. Search on ‘recommendation engines’. Has been a topic of research since decades.

Please try Spotify as well, and use the MoodAgent plugin.

There are so many API’s around. Would be cool to hook into these calls.

I think I’d have to work hard with Musictivate!

Well, I guess I can follow on what you’ve suggested, Hunt.

And, Spotify, just like Google Music, seems to not have been launched in India yet. But the website gives ideas about the application.

Thanks for the continuous help; I’d be reporting as I go with the ideas!