Acoustic scene classification (ASC) is important for context-aware applications to recognise the environment based on the available acoustic information. Despite its promising application prospects, ASC is a challenging problem due to both the similar characteristics of some scenes and complexity of sounds present in other scenes. This dissertation presents findings in the field of ASC utilising hand-crafted visually inspired features extracted from a 2D time-frequency image representation. The time-frequency visual representations of the temporal and spectral structures for each acoustic scene with suitable feature extraction methods were shown to enhance the identification of the different scene classes.
|Qualification||Doctor of Philosophy|
|Award date||30 Oct 2019|
|Publication status||Unpublished - 2019|