Unoffical empeg BBS

Quick Links: Empeg FAQ | RioCar.Org | Hijack | BigDisk Builder | jEmplode | emphatic
Repairs: Repairs

Topic Options
#25860 - 31/01/2001 21:55 Self learning dj
gbeer
carpal tunnel

Registered: 17/12/2000
Posts: 2665
Loc: Manteca, California
Sort of like teaching match boxes to play tick-tac-toe.

The player starts off with a blank slate playing randomly. As each song is played, the user rates the seguey. This is recorded in the DB. Later this is used to bias the random play away from bad combinations.

Glenn

Yeah, I know seguey is spelled wrong. But neither netscape-mail or look could help me.



_________________________
Glenn

Top
#25861 - 01/02/2001 03:24 Re: Self learning dj [Re: gbeer]
peter
carpal tunnel

Registered: 13/07/2000
Posts: 4180
Loc: Cambridge, England
10,000 songs = 100,000,000 segues.

It'll never make the same segue twice anyway (or, at least, it'll take on average 1,141 years before a full player does).

And no, I've got no idea why a word pronounced segway is spelt "segue". Is it Spanish? Is there an accent missing?

Peter



Top
#25862 - 01/02/2001 09:55 Re: Self learning dj [Re: peter]
schofiel
carpal tunnel

Registered: 25/06/1999
Posts: 2993
Loc: Wareham, Dorset, UK
It's French, and yes, there's a grave accent missing from the last E.

One of the few remaining Mk1 owners... #00015
_________________________
One of the few remaining Mk1 owners... #00015

Top
#25863 - 02/02/2001 03:26 Re: Self learning dj [Re: peter]
smu
old hand

Registered: 30/07/2000
Posts: 879
Loc: Germany (Ruhrgebiet)
Hi Peter.

You are assuming that
a) the shuffle function works perfectly random
b) the like/dislike sequence matrix is completely populated
While (a) might be the case some day in the future (but certainly not for 1.02), (b) will never be the case. However, I agree with you that such a rating system in the player itself wouldnīt be worth the hassle. However, with an appropriate edit function in emplode, this would be great, because one could fill the matrix much more systematically. Furthermore, the empeg (player) would never have to read the complete matrix, just the list of songs the listening person would like/dislike to follow the song that was just added.
Regarding the imperfect shuffle that the player currently uses:
I noticed that a specific song is usually found within 20-30 positions from its unshuffled position in the playlist. The reason why I checked this was that I have all Meatloaf CDs on my unit, with the "Live at Wembley" album being a combination of "RockīnīRoll hero" and some other album (probably "Bat out of Hell II" or "Welcome to the neighbourhood"). Now I got two pairs in a row of a song from the live album and the same song from the normal album. Not too funny, but only two songs later, the same thing happened again (another pair of live/normal versions of the same song). After this incident, I checkd the playlist in its sorted/shuffled form (having the fun to write down the song titles and origins by hand for about 80 for the Meatloaf songs, and about the first 120 of the "complete set" of about 500). I found the above rule of an average move of about 10-20 positions, that only got broke by about 3 songs of the meatloaf playlist and not more than 10 of the 120 of the complete playlist.
Now I would suggest the following:

  • When shuffling a playlist, try to enhance the shuffling function itself, so that the songs move by at least 30-40 songs in average (if the playlist is large enough, probably should be at least 25-40% of the total playlist length).
  • Check the resulting (shuffled) playlist, and try to eliminate sequences where two songs of similar title (first x characters match, or one title can be found completely in the other songs title, eg. "Bat out of Hell" and "(Live) Bat out of Hell" or "Bat out off Hell (Live)") are following each other (or even better: arenīt divided by at least two other songs).


cu,
sven
(MkII 12GB blue now green, #080000113)
_________________________
proud owner of MkII 40GB & MkIIa 60GB both lit by God and HiJacked by Lord

Top
#25864 - 02/02/2001 03:30 Re: Self learning dj [Re: smu]
peter
carpal tunnel

Registered: 13/07/2000
Posts: 4180
Loc: Cambridge, England
Check the resulting (shuffled) playlist, and try to eliminate sequences where two songs of similar title (first x characters match, or one title can be found completely in the other songs title, eg. "Bat out of Hell" and "(Live) Bat out of Hell" or "Bat out off Hell (Live)") are following each other (or even better: arenīt divided by at least two other songs).

That's quite a good idea... I'll have a think.

Oh, and I checked dictionary.com. It's Italian.

Peter



Top
#25865 - 03/02/2001 09:15 Re: Self learning dj [Re: gbeer]
94cobra
enthusiast

Registered: 30/09/1999
Posts: 252
I think ths would be great and should not be to difficult. All you would need is a list of songs that were 'selected' rather than every song played. Don't count the songs played in shuffle mode. The list would keep track of which songs have been played the most an play those back first, then when it ran out of it's own list it could continue to play the rest of the songs on the player. Kinda like a recent documents setting in Windows.

Or have I missed the point here?


Proud Owner of MK2 080000558 - 18gb Blue
_________________________
Sonic Blue 03 Cobra Vert Owner!!!

Top