Current students


Section: Telecommunications

Major Research topic:
Merging machine learning and signal processing methods for audio applications.

My research project focuses on the possibilities offered by machine learning techniques in audio analysis applications. My goal is to investigate the possibility of merging classic signal processing techniques with more recent feature learning architectures and classifiers. In particular, usually a model-based approach should be chosen for designing a significant representation of the audio signals, while a data-driven approach should be employed for learning a solution to problems where modelling is unfeasible.  I am planning to apply the proposed methodology to two research fields: Music Information Retrieval and Audio Forensics. In the field of Music Information Retrieval my goal is mainly to extract high level semantic information from musical signals. This research field is moving toward the use of new deep learning architectures for feature learning and prediction, given the intrinsic variable nature of music signal and the impossibility of a priori modelling concepts related to music perception and enjoyment.On the other side, the field of Audio Forensics, in particular the detection of splicing, manipulation or editing for authenticity verification, often works with speech signals. The nature and the predictability of speech allow to work with model-based features, specifically tailored for the application, in combination with more traditional classification systems. The duality of these two research fields allow me to explore several solutions and possibly try to merge the two approaches.