My research project
Recognition of paralinguistic features for singing voice description
For high quality singing, performers need to listen to aspects of their singing voice to hear and control vocal features and techniques such as breathy, pressed or vibrato singing: these are known as paralinguistic features. The aim of this project is to use machine learning methods to automatically recognize a range of paralinguistic features of singing voice, to be able to describe these in a way that would correspond to an expert human listener. This research is likely to build on deep learning methods for audio feature analysis and audio source separation, using either hand-crafted features or representation learning.