Project information
Sound source localization is a fundamental task, especially in remembrance and multiple sources
environments; it includes recognizing the temporal onset and offset of sound events when active,
classifying the sound events into a known set of classes, and further localizing the events in space when
active using their direction of arrival (DOA). In this project, we work with 3D audio sounds captured by a
first-order Ambisonic microphone and these sounds are then represented by spherical harmonics
decomposition in the quaternion domain. The project aims to detect a known set of sound event classes'
temporal activities and locate them in the space further using quaternion-valued data processing. In
particular, we focus on sound event localization and detection (SELD). To do this, we use a given
Quaternion Convolutional Neural Network with the addition of some recurrent layers (QCRNN) for the joint
3D sound event localization and detection task.
