Primjena strojnog učenja u klasifikaciji zvučnih signala

master's thesis

• Title (english): Application of machine learning in classification of sound signals
• Author: Mario Štrbac
• Mentor: Antonio Petošić (mentor)
• Committee member: Antonio Petošić (predsjednik povjerenstva)
• Committee member: Ivan Đurek (član povjerenstva)
• Committee member: Marko Horvat (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2020-07-06, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Ovaj rad bavi se problemom klasifikacije zvučnih signala iz okoliša primjenom strojnog učenja. Za skup podataka odabran je skup gradskih zvučnih isječaka UrbanSound8K. Klasifikacija zvuka testirana je pomoću dva pristupa dubokog učenja u programskom jeziku Python. Prvi pristup uključuje konvolucijsku neuronsku mrežu implementiranu s Keras bibliotekom, a drugi metodu prijenosnog učenja ostvarenu pomoću modela ResNet-18 i biblioteke fastai. Za reprezentaciju zvuka odabran je spektrogram koji svojom vremensko-frekvencijskom karakteristikom može dati dobar uvid u svojstva okolišnih zvučnih signala. Konačno, napravljena je analiza dobivenih rezultata i usporedba oba modela te su predloženi dodatni koraci za njihovo moguće poboljšanje.

Abstract (english)

This paper examines environmental sound classification using machine learning methods. The methods are applied on UrbanSound8K data set consisting of urban sound excerpts. For sound classification two deep learning methods are tested in Python programming language. First approach includes convolutional neural network implemented with Keras library and the second refers to transfer learning implemented with ResNet-18 model and fastai library. Spectrograms were used for sound representation because of their time-frequency characteristic and suitability in describing environmental sound features. Finally, analysis of results is given, as well as comparison of both models with additional steps proposed for further improvements.

• Keywords: klasifikacija okolišnih zvukova; spektrogram; strojno učenje; konvolucijske neuronske mreže; prijenosno učenje; UrbanSound8K
• Keywords (english): environmental sound classification; spectrogram; machine learning; convolutional neural networks; transfer learning; UrbanSound8K
• Language: croatian
• URN:NBN: urn:nbn:hr:168:394700
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-13 14:00:21