Време29. јун 2012. 15:32
ПредавачНаташа Рељин, Delaware State University, Applied Mathematics Research Center
Местосала 61, Електротехничког факултета

IEEE S&M Section, CAS-SP joint Chapter и
Електротехнички факултет, Универзитета у Београду организују предавање:


Sala 61, petak, 06.07.2012, 18:00


The amount of multimedia content which can be found on the Internet is growing immensely, every day. The retrieval of audio information becomes more exhausting and time consuming. In addition, not every such information has a text label nor it is provided necessarily in the same language. There are several problems rising at this point, such as: how to identify the unknown melody (song), how to identify the unknown performer, how to find different melodies of the same performer or same melodies of different performers, how to distinct different sounds of the car engine, how to distinct heart sounds,… It is important to develop an effective way of retrieving the material of interest. The content of multimedia information should be represented with as fewest as possible features, which will describe details of the signal, as well as the general form of it, in the most accurate way, forming the feature vectors. This way the classification of melodies could be done by comparing their feature vectors.

In this presentation, some of the methods used for classification of the audio signals are presented. Two types of feature sets will be presented:  mel-frequency cepstral coefficients (MFCC), and characteristic values obtained from the multifractal analysis. As classification tool, support vector machines (SVM) will be used. The classification accuracy will be calculated for each set of the features and presented.

Predavač: Natasa Reljin

Delaware State University, Applied Mathematics Research Center, Dover, Delaware (USA)


Natasa Reljin graduated from the School of Electrical Engineering, University of Belgrade, in 2005, at the Department of Telecommunications. As of 2006, she is a PhD student at the Delaware State University, Dover, USA, at the Department of Applied Mathematics and Theoretical Physics, where she also works as Teaching and Research Assistant. She published a number of conference papers in the fields of image and audio signal processing and data mining, as well as book chapter in adaptive filtering. In 2008 she won the first place in a student research competition in New Orleans. In 2010 and 2011 she was reviewer for TELFOR conferences.

Ms. Reljin was involved in various projects, such as: EU COST-292 project “Semantic Multimodal Analysis of Digital Media”, and different US projects such as: Transportation and Public Safety Information Exchange Pilot Project, Adaptive Approach to Identify Unusual Behavior in Video Imagery, Creation of the Center for Research and Education in Optical Sciences and Applications (CREOSA) at Delaware State University, Applied Mathematics Research Center.

Miroslav Lutovac