[vlc-devel] Idea for GSoC 2010

Guy Perkins me at atdotmedotuk.me.uk
Mon Mar 22 02:22:50 CET 2010

Interesting, I would also adopt a Mood points out of 10 click for the
user on tracks played after the intitial, with another button for "pick
another/reject", & then capture & use that data to refine/analyse,
databased of course. Thus capturing & identifying attribute anomolies
that may not be initially detected within the current filters, allowing
for further refinement.
Personally, I believe that this type of facillity is the future, as it
expands musical horizons. (Let's face it, a lot of users these days have
let's say 20K tracks, but only have really listened to 7-8K)

-----Original Message-----
From: vlc-devel-bounces at videolan.org
[mailto:vlc-devel-bounces at videolan.org] On Behalf Of Pranav Agarwal
Sent: 22 March 2010 01:02
To: vlc-devel at videolan.org
Subject: [vlc-devel] Idea for GSoC 2010

Hello all,

Im Pranav Agarwal currently pursuing MSc Maths & BE Computer Science
from BITS, Pilani, India. I have an idea which i'd like to implement on
the VLC Media Player as a part of the GSoC 2010.


The objective is to develop an application that creates a playlist of
audio/video files based on User's Mood and the first song chosen to
play, by extracting certain attributes of music files automatically
through Signal Classification techniques or an Online metadata search


Say a user has 100 songs in a directory that he adds to the media player
as a potential playlist. Assuming the directory has songs of different
Genre, a user might not be interested to listen to all the songs but a
few which suits his mood. The application does so by finding the
similarity between the songs on the basis of several features like
frequency spectrum, zero-crossing rate, tempo, beat strength etc through
Signal Processing techniques and help identify a potential playlist for
the user depending on the mood.
Moreover, the application depends on what according to user is Rock,
Classical or any other genre as according to one particular user, the
song could be a soft rock and according to someone else it could be a
classical one and since, there is no standard way of finding the Genre
of the song, User Perception plays an important role in categorizing
songs. This kind of a plugin to VLC could be added as another
mode(something similar to the Random Mode).

The first part of the project involves using the Signal Processing
techniques to find the attributes such as the spectral centroid, zero
crossing rate(both mean and standard deviation), beats per minute, beat
strength(both mean and standard deviation), tempo to find the hidden
attributes of the song and thereby calculate its distance from the other
songs in the list. By putting in a threshold, we can put the songs in an
order and thereby create an playlist depending upon the users's mood.
The second part will be to integrate it with the VLC Media PLayer.

What I've done till now?

I have presently written the code for finding the attributes (as
mentioned above) for any particular song in Matlab. I'd like to extend
my project so that it gets implemented on VLC and thereby relieve user's
of constantly clicking on a song from a list of songs.
the project that i have implememted has one limitation which is
"extracting the attributes for a mp3 file with size greater than 4MB."
I'm currently working on this part.

I intend to participate in GSoC and work with the VLC community and hope
to implement this feature on VLC. My programming skills are C, Ruby,
Shell Scripting. If this idea looks good, i'd like to know, what are the
programmable skills required?

Waiting for a reply, so that someone can guide me through as to how can
i go about working with it and take up this idea further. 


Pranav Agarwal
meetpranav123 at gmail.com

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/vlc-devel/attachments/20100322/5ebcb714/attachment.html>

More information about the vlc-devel mailing list