Thesis Details

Recognition of Audio Events Using Deep Neural Networks

Bachelor's Thesis Student: Uchytil Albert Academic Year: 2015/2016 Supervisor: Schwarz Petr, Ing., Ph.D.

Czech title

Language

English

Abstract

A lot of information is carried in sound. The amount of audio data is increasing with a growing technical level of the society. With more data, the task of processing it gets harder for human beings. This thesis is about recognition of audio events using neural networks. We focused on classification of phonemes and their categories. We used the Multilayer perceptron model as a classifier. We examined the relation between the accuracy of the model and its properties. Our goal was to estimate the network setup to obtain the best results. The accuracy is influenced by input features. We examine the relation between a type of the features and the success rate. The differences between input feature types are reduced by using the context. The bigger context we use the better results we get. Problem is, when contexts overlap, overlapping leads to a higher error rate. We have used a neural network with three hidden layers.

Keywords

Sound recognition, Audio classification, Neural Networks, Phoneme classification

Department

Department of Computer Graphics and Multimedia FIT BUT

Degree Programme

Information Technology

Files

Status

defended, grade C

Date

15 June 2016

Reviewer

Černocký Jan, prof. Dr. Ing.

Committee

Černocký Jan, prof. Dr. Ing. (DCGM FIT BUT), předseda
Bidlo Michal, doc. Ing., Ph.D. (DCSY FIT BUT), člen
Drahanský Martin, prof. Ing., Dipl.-Ing., Ph.D. (DITS FIT BUT), člen
Rychlý Marek, RNDr., Ph.D. (DIFS FIT BUT), člen
Španěl Michal, Ing., Ph.D. (DCGM FIT BUT), člen

Citation

UCHYTIL, Albert. Recognition of Audio Events Using Deep Neural Networks. Brno, 2016. Bachelor's Thesis. Brno University of Technology, Faculty of Information Technology. 2016-06-15. Supervised by Schwarz Petr. Available from: https://www.fit.vut.cz/study/thesis/18850/

BibTeX

@bachelorsthesis{FITBT18850,
    author = "Albert Uchytil",
    type = "Bachelor's thesis",
    title = "Recognition of Audio Events Using Deep Neural Networks",
    school = "Brno University of Technology, Faculty of Information Technology",
    year = 2016,
    location = "Brno, CZ",
    language = "english",
    url = "https://www.fit.vut.cz/study/thesis/18850/"
}

Theses