Evaluating performance of shuffling data augmentation techniques for audio event detection

Kelly, David Leslie (2022) Evaluating performance of shuffling data augmentation techniques for audio event detection. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (868kB) | Preview
Preview	PDF (Configuration manual) Download (711kB) | Preview

Abstract

Audio Event Detection is an emergent field of machine learning. The goal of is to the world around us through sound. Animal habitat preservation, ambient assisted living and preventative machine maintenance are all fields exploring the commercial use of AED to augment human decision machine. A key challenge stalling the development of AED systems is the lack of high-quality audio data labelled audio data. Producing such data is an expensive and time-consuming task. Techniques which offer a classification performance uplift in data constrained domains are of particular interest to AED research. This technical report describes the design and replication of the best performant submission from DCASE 2018 Task 5. The research critically evaluates the technique proposed by Inoue. Through experimentation, a set of alternatives of experimental parameters were discovered which exceed the performance of the reference implementation on the same challenge dataset. This research shows promise for the field of audio event detection where the use of mitigation to deal with a lack of high-quality labelled data for a given classification scenario is common. This work also shows an improvement in performance over the baseline work when comparing GPU compute effort required to reach equivalent performance classification performance.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Sahni, Vikas UNSPECIFIED
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Cloud computing Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Cloud Computing
Depositing User:	Tamara Malone
Date Deposited:	18 Apr 2023 18:12
Last Modified:	18 Apr 2023 18:12
URI:	https://norma.ncirl.ie/id/eprint/6473

Actions (login required)

View Item