The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9766

Title:Automatic Inference of Humpback Whalesong Grammar
Date:Sep 1997
Abstract:Humpback whales emit a complex, structured sequence of sounds known as a song; the study of hujpback whalesong aids understanding of their lifestyle. the song is built up from 'Units' taken from a small set of classes; the reliable classification of these units is of paramount importance in studying the song. The following study examines a system for extracting and categorising humpback whalesong units. A two layer hierarchy of self-organising Kohonen maps is used to perform coarse and fine categorisation of units into classes; labelling of songs can be produced. The system extends previous work carried out in the Department of Artificial Intelligence in the summer of 1996; units are now extracted and represented more accurately. The two layers of maps were found to be successful in classifying units, although the degree of success is difficult to evaluate in the absence of objective criteria. Analysis of different resolutions for the system was undertaken.

[Search These Pages] [DAI Home Page] [Comment]