Speech Recognition

Authors

Dimitrije Pešić, First Gymnasium, Kragujevac

Lazar Zubović, Gymnasium, SOmbor

Mentor

Pavle Pađin, School of Electrical Engineering, University of Belgrade

Nataša Jovanović, School of Electrical Engineering, University of Belgrade

Abstract

Speech recognition is one of the biggest challenges of technology. The growing need for digitalization is followed by the need to expand knowledge in this field. Research so far shows the effectiveness and accuracy of speech recognition methods with or without deep learning. This paper focuses on observing and comparing various methods such as convolutional nerual networks and data classifiers that don’t use deep learning in order to determine the best approach for identifying words. Testing on the FSDD word database and a database consisting of Serbian words, it was determined that the most accurate way to process audio recordings is by using convolutional neural networks, so it is most optimal to conduct further research in that direction.

Full Paper

For the complete technical details, methodology, and results, please refer to the full paper in Serbian.