Für eine korrekte Darstellung dieser Seite benötigen Sie einen XHTML-standardkonformen Browser, der die Darstellung von CSS-Dateien zulässt.


Content Based Audion Retrieval Methods. Cases of Violin Sound Modeling and Musical Genre Classification

Dr. -Ing. Ewa Łukasik, Poznań University of Technology, Institute of Computing Science, Poland

Date: 13/11/2014
Time: 10:15
Room: H-F 001

Content based audio retrieval  is an interdisciplinary research domain that focuses on the automatic extraction of information from audio signals. It enables searching the indexed information for different purposes, such as automatic speech recognition, audio forensics, music information retrieval, audio segmentation, environmental sound retrieval, audio surveillance or audio based/supported  video retrieval. This talk will first concentrate on the specificity of content-based audio features.  A challenge is to choose  appropriate features for the representation of specific audio signals. Then the problem of the identification of perceptually similar audio content will be discussed: the same sound may be assigned by humans to different semantic categories and, on the other hand, very similar sounds may be well distinguished by experts.
The above topics will be illustrated by two cases. First relates to the musical genre recognition. Second is devoted to the modeling similarly sounding musical instruments - violins - competing during the violinmaking contest. In particular a problem of inducing jury’s preferences in terms of acoustic features of violin sounds will be presented. The preference model explains the jury’s ranking and permits to build a ranking of a new set of violins according to this policy.