Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.781008
Title: Automatic detection and classification of bird sounds in low-resource wildlife audio datasets
Author: Morfi, G.
ISNI:       0000 0004 7966 6445
Awarding Body: Queen Mary University of London
Current Institution: Queen Mary, University of London
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
There are many potential applications of automatic species detection and classification of birds from their sounds (e.g. ecological research, biodiversity monitoring, archival). However, acquiring adequately labelled large-scale and longitudinal data remains a major challenge, especially for species-rich remote areas as well as taxa that require expert input for identification. So far, monitoring of avian populations has been performed via manual surveying, sometimes even including the help of volunteers due to the challenging scales of the data. In recent decades, there is an increasing amount of ecological audio datasets that have tags assigned to them to indicate the presence or not of a specific bird species. However, automated species vocalization detection and identification is a challenging task. There is a high diversity of animal vocalisations, both in the types of the basic syllables and in the way they are combined. Also, there is noise present in most habitats, and many bird communities contain multiple bird species that can potentially have overlapping vocalisations. In recent years, machine learning has experienced a strong growth, due to increased dataset sizes and computational power, and to advances in deep learning methods that can learn to make predictions in extremely nonlinear problem settings. However, in training a deep learning system to perform automatic detection and audio tagging of wildlife bird sound scenes, two problems often arise. Firstly, even with the increased amount of audio datasets, most publicly available datasets are weakly labelled, having only a list of events present in each recording without any temporal information for training. Secondly, in practice it is difficult to collect enough samples for most classes of interest. These problems are particularly pressing for wildlife audio but also occur in many other scenarios. In this thesis, we investigate and propose methods to perform audio event detection and classification on wildlife bird sound scenes and other low-resource audio datasets, such as methods based on image processing and deep learning. We extend deep learning methods for weakly labelled data in a multi-instance learning and multi task learning setting. We evaluate these methods for simultaneously detecting and classifying large numbers of sound types in audio recorded in the wild and other low resource audio datasets.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.781008  DOI: Not available
Keywords: bird sounds ; automatic species detection ; avian population monitoring ; audio event detection
Share: