Digg meets LinkedIn (a free* idea)
LinkedIn is a social network, allowing people who know someone in common to find each other.
It shouldn't be too hard to swap people for concepts. It could work well with music (because musical artists are people, but it isn't the person, but rather the product).
So, when you join the site, you enter your 10 favorite artists (too many artists would create too many matches). You can then immediately find the profiles of anyone that has 3 or more of the same artists (and ask them to be your friend etc).
The Digg aspect kicks in with songs rather than news. And, like news, the songs have to be fresh.
The artist of a new song gets to nominate 5 artists who they feel akin to. Any user who has one of those as their 10 favorite artists get to listen to the new song (yes, the artist could select the most popular 5 artists, but those people would be so innundated with new songs they'd actually listen to few of them).
When you listen to a new song, you get to rate it, and those users who have some of the same 10 favorite artists as you get the song recommended to them. The closer their 10 favorite artists are to yours, the more strongly the song gets recommended by you.
So, basically, if someone who has 6 of my 10 favorite artists in their 10 favorite artists, loves a song and rates it highly - I will have it strongly recommended to me.
But if someone who has 1 of my 10 favorite artists in their 10 favorite artists, loves a song and rates it highly - I will have it recommended to me, but not strongly - it might appear at the bottom of 200 recommended tracks that week.
You could even give weighting to the opinions of others - that way famous folk could get involved, and I could choose to give top weighting to any song that Victoria Beckham liked (as long as she has 1 of her 10 favorite artists the same as mine).