Audio Mining with emphasis on Music Genre Classification

Anders Meng

AbstractAudio is an important part of our daily life, basically it increases
our impression of the world around us whether this is communication,
music, danger detection etc.
Currently the field of Audio Mining, which here includes areas of
music genre, music recognition / retrieval, playlist generation
etc. is receiving quite a lot of attention. The first breakthough in
audio mining was created by MuscleFish in 1996, a simple audio
retrieval system. With the increasing amount of audio material being
accessible through the web, e.g. Apple's iTunes (700,000+ songs),
Sony, Amazon, new methods in searching / retrieving audio effectively
is needed. Currently, search engines such as e.g. Google, AltaVista
etc. do not search into audio files, but uses either the textual
information attached to the audio file or the textual information
around the audio. Also in the hearing aid industries around the world
the problem of detecting environments from the input audio is
researched as to increase the life quality of
hearing-impaired. Basically there is a lot of work within the field of
audio mining.

The presentation will mainly focus on music genre classification where
we have a fixed amount of genres to choose from. Basically every audio
mining system is more or less consisting of the same stages as for the
music genre setting. My research so far has mainly focussed on finding
relevant features for music genre classification living at different
timescales using early and late information fusion. It has been found
that for the task of music genre classification, the features, and
their temporal relationships are very important when determining the
music genre.
KeywordsAudio Mining, Feature extraction, Classification, Eearly / Late information fusion
TypeMisc [Presentation]
Year2004
NoteA presentation as to introduce some of my work at IMM to the ISIS group in Southampton.
Electronic version(s)[pdf] [zip]
BibTeX data [bibtex]
IMM Group(s)Intelligent Signal Processing


Back  ::  IMM Publications