Exclude tracks that are too similar to the seed tracks

[GoogleCodeExporter]

Mirage should exclude tracks that are so similar as to be considered
identical to the seeds.

A user might have duplicate tracks in his library, or several occurence of
the same song (from an album and the single, for example). He doesn't want
to hear the same song two times in a row.

Original issue reported on code.google.com by bertrand.lorentz@gmail.com on 8 Aug 2008 at 3:28

[GoogleCodeExporter]

I'll change the Similarity Computation function to return values between 0 and 
1.

But i think it will be difficult to find a 'correct' similarity value to 
identify
'almost identical' songs.

Original comment by dominik....@gmail.com on 4 Nov 2008 at 9:22

[GoogleCodeExporter]

It might me helpful in this case also to consult the track information rather 
than
relying entirely on the songs' similarity. I am not good with the algorithms, 
but I
have the impression that different versions of the same song result in "over 
fitting"
recommendations.

Original comment by cpgl...@gmail.com on 7 Dec 2008 at 11:32

[GoogleCodeExporter]

The is the patch discussed at
http://groups.google.com/group/mirage-list/browse_thread/thread/4c74f51f05fe0964

Original comment by wademen...@gmail.com on 14 Feb 2009 at 1:26

Attachments:

• dupefilter.patch

[GoogleCodeExporter]

I committed a first step to this, inspired by Wade's patch :
Mir.SimilarTracks can now ignore tracks whose distance to the seeds is below a
certain threshold. For now, those tracks are not remembered, and the tracks 
added to
the playlist are not compared to each other.

As a bonus, the number of tracks added to the playlist and the value of the 
threshold
are now configurable. You can set their values through gconf.

Original comment by bertrand.lorentz@gmail.com on 14 Feb 2009 at 4:13

[GoogleCodeExporter]

We could add a (Right-Click) preferences Item to the "Playlist Generator" item.

Possible configurable items: 
[ ] filter too similar songs
 * Use Aritst Filter (max n same-artists-tracks in playlist)
 * Use Album Filter
 * Similarity Threshold

Do you think this would be a good idea to way to handle the too-similar songs 
that
way...?

Original comment by dominik....@gmail.com on 7 Mar 2009 at 6:33

[GoogleCodeExporter]

Those preferences would be nice indeed.

A nice change has been recently committed to banshee SVN that might help with 
this :
sources can now easily add their preferences to the main preferences dialog.
You can take a look at this :
http://banshee-project.org/~gburt/tmp/Screenshot-Preferences.png
Audio CD support is an extension, so it should demonstrate how to do this.

Original comment by bertrand.lorentz@gmail.com on 8 Mar 2009 at 9:04

[GoogleCodeExporter]

Issue 51 has been merged into this issue.

Original comment by bertrand.lorentz@gmail.com on 2 Feb 2010 at 10:13

[GoogleCodeExporter]

http://rudd-o.com/new-projects/python-audioprocessing youllfind code to detect 
duplicates there.

Original comment by dragonf...@gmail.com on 15 Aug 2010 at 8:46