[Banshee-List] Mirage - Automatic Playlist Generation Plugin
- From: Dominik Schnitzer <dominik schnitzer at>
- To: banshee-list gnome org
- Subject: [Banshee-List] Mirage - Automatic Playlist Generation Plugin
- Date: Fri, 31 Aug 2007 16:21:12 +0200
Hi,
I wanted to wait and polish a bit more before publishing this, but since
this already found its way on planet.gnome.org... :-)
I've written an automatic playlist generation plugin for Banshee
(Mirage). This was done as a part of my master thesis. Once the plugin
is activated in Banshee it "listens" to your music collection (this can
take quite long...) and tries to compute a timbre-similarity model of
it. Those models are stored in an sqlite database. (This only needs to
be done once)
After all your songs are analyzed. You can generate playlists by
dragging a seed song on the Playlist Generator item in banshee. Mirage
then searches its database for similar sounding songs. The 5 top-hits
are then put in the playlist.
This is useful in large collections, where random shuffle fails.
Have a look at the video to see how automatic playlist generation works.
Homepage: http://hop.at/mirage/
(includes tarballs, install instructions, ubuntu feisty .deb, a
screenshot && video)
Note:
* If you listen to more than 60% of one of the songs in the generated
playlist - the playlist updates itself by searching for similar songs of
this song.
* if you skip songs in the playlist, those song is excluded from
further playlists
* MPG123/FAAD/OGG123/SOX are reguired because of performance reasons.
Since Mirage needs to decode every music piece in order to analyze it
the fastest decoders available are needed. I.e. using the gstreamer
framework would have been really nice, but in the end it was too slow
:-/ Even libmad is about 4-5 times slower than MPG123, just to name an
example. Opinions on this issue are welcome!
It would be very cool, if someone could try this out and give some
feedback. I know it is far from being perfect but hey, it somehow works
:). Patches are welcome. I am planning to improve and work on this,
since smooth integration in Banshee and functionality like this would be
cool.
greetings,
domi.
PS: I'm currently working at the Department of Computational Perception
at the Johannes Kelpler University (Austria, Linz, http://ww.cp.jku.at).
Our department is heavily working in the field of music similarity,
music information retrieval and automatic playlist generation.
[
Date Prev][
Date Next] [
Thread Prev][
Thread Next]
[
Thread Index]
[
Date Index]
[
Author Index]