Problems with short tracks?

0 views
Skip to first unread message

eric casteleijn

unread,
Nov 12, 2008, 11:50:10 AM11/12/08
to mirag...@googlegroups.com
First of all: congrats on getting 0.4 shipped, I look forward to
updating my python implementation to reap the benefits.

I finally have mirage fully working from within my QL plugin, and the
results are pretty awesome. I decided to use a rather arbitrary cut-off
point for the maximum distance I still consider 'similar' and 70 seems
to be working nicely for me, or at least I haven't seen any
freakommendations yet. :)

Now for my question: The analysis seems to break when I pass it a short
track (I think things under a minute break pretty consistently) Is this
known, or is it tied to my implementation, or is it perhaps something
that went away in 0.4?

For now I'm skipping songs that are shorter than a minute.

--
- eric casteleijn
http://thisfred.blogspot.com

Dominik Schnitzer

unread,
Nov 12, 2008, 11:59:40 AM11/12/08
to mirag...@googlegroups.com
Hi,

I'm running genre classification-evaluations with 30s snippets of
tracks with Mirage 0.4, so I guess short tracks should not be a
problem any more. (?)

But in general, I'd say skipping tracks shorter than a minute is
reasonable anyways.. Btw, the value "70" where you cut off playlists
now should be around 15, since the similarity function changed..

greetings,
dominik.

--
DI Dominik Schnitzer
Austrian Research Inst. for Artificial Intelligence
Freyung 6/6, A 1010 Vienna, Austria, Europe
Fax: +43-1-5336112-77(Fax)

Reply all
Reply to author
Forward
0 new messages