W3C home > Mailing lists > Public > public-xg-mmsem@w3.org > July 2006

Music use case

From: Giovanni Tummarello <g.tummarello@gmail.com>
Date: Wed, 19 Jul 2006 19:09:47 +0200
Message-ID: <44BE675B.9090602@gmail.com>
To: MMSem public <public-xg-mmsem@w3.org>
Attached the music use case named Social Semantic Browsing.
We'll really appreciate feedbacks on this scenario. Yell now or help 
afterwards! :-)
Giovanni



----

*Music related use case: social semantic browsing*

*Authors: Giovanni Tummarello, Christian Morbidoni*


A great number of Internet users have consistent music collection 
composed of locally playable files in formats such as MP3, Ogg Vorbis etc..

Such files often come from diverse sources such as online shops, file 
sharing, direct CD ripping, personal recording etc.

We assume that such users are connected to the Internet and therefore 
can access both the “semantic web” and specialized services.

As the number of such media is likely to be great, “sensible” browsing 
using traditional methods (e.g. Picking from a list) is not possible. By 
“sensible” browsing we mean selecting a “next” piece according to 
characteristic of the “previous” so that, for example, there might be 
matches or not strong mismatches in genre, author, timber, rhythm, 
velocity, instruments, acoustic pressure etc.

In order to assist the user (or automatically act, e.g. Compose a 
playlist, upon such premises) both low level features and higher level 
metadata are needed.

The metadata associated with such audio collections files are 
heterogeneous in quality, if they exist at all.

Metadata associated with such files are usually limited to file names, 
file system informations (e.g. The names of the folder in which the user 
has saved the files) and ID3 like information headers. Except for fields 
like “genre” which is encoded as integer, such information fields are 
typically encoded as strings which do not, naturally, act as unique 
identifiers.

To address this use case, there is need of using a combination of 
techniques such as text heuristics combined with low level audio 
features algorithms to provide fast and reliable identification of the 
tracks, low level audio features algorithms to extract content 
descriptions and social semantic web technologies, e.g. Semantic web P2P 
to retrieve high level metadata once the tracks have been identified.




Received on Wednesday, 19 July 2006 17:10:31 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Tuesday, 8 January 2008 14:21:19 GMT