Abstract
dc:description.abstractWhen conducting passive acoustic monitoring of humpback whale songs in St Marie channel, Madagascar, sometimes recordings containing multiple singers were obtained. In this case, separating the mixtures and obtaining a recording of an individual singer is of interest. The specific method that we utilized for source separation is adapted from the proposal by Sawada et al. This algorithm can effectively operate in conditions with strong reverberation. It can also potentially cope with underdetermined mixtures where source number exceeds hydrophone number.<br/><br/>The effectiveness of the Sawada method is verified through separation of artificial humpback song mixtures generated by the impulse responses of underwater channel model. However, this method is unreliable for the separation of real humpback whale songs as a consequence of severe background noise. We propose a noise reduction method based on weighted median threshold scheme, which significantly improves source separation performance of real recording in severe noise environments.<br/><br/>As to the Sawada algorithm, the number of sources need to be known in order to conduct source separation. However, in reality, the number of source is unknown, Hence, we need to estimate it before performing sources separation. Various methods for automatically estimating the number of sources are investigated in this thesis, and the units counting method turns out to be the most promising one.
Degree
thesis:*- Name dc:type.qualificationname
- Ph.D.
- Level dc:type.qualificationlevel
- doctoral
- Grantor dc:publisher.institution
- University of Southampton
- Year dc:date.issued
- 2016
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Zhang, Zhenbin
- Advisor dc:contributor.advisor
-
- White, Paul
- ORCID person:identifier.orcid
Rights
- Language dc:language
- English, eng