Acoustic event detection aims to detect the event class and to localize the start and the end times of the audio events in various real world scenarios. The audio events are either monophonic or polyphonic depending on whether they are isolated or overlapping with each other. This thesis proposes deep learning solutions utilizing both the event class and event localization information for improving the detection performance when dealing with monophonic or polyphonic sound events. A data augmentation solution is also proposed that addresses limitations in the available training data when deploying deep learning for acoustic event detection.
|Qualification||Doctor of Philosophy|
|Award date||1 Jul 2019|
|Publication status||Unpublished - 2019|