Current students


Section: Telecommunications

Major Research topic:
High-Order Soundfield Processing: Novel Modeling Paradigms for Acoustics and Plenacoustics

Soundfield analysis, processing and rendering have been steadily growing in popularity for several decades. In fact, the ever-evolving trends in immersive multimedia communication, have triggered a rapid evolution of computational acoustics from the early solutions of spatial audio processing, to the emerging paradigms of interactive extended realities.

Soundfields are usually captured and rendered with multiple microphones and loudspeakers, usually arranged in clusters or arrays. Simple arrangements of transducers of this sort are already integrated in commercial devices such as smart TVs and smart speakers. However, classical space-time processing techniques tend to rely on strong modeling assumptions that might limit the accuracy and flexibility of such solutions in specific applications, particularly when strong requirements of low-invasivity and interactivity are in place.

Recently, the Ray Space Transform (RST) has been proposed for efficiently capturing and flexibly processing the plenacoustic image of an acoustic scene. This corresponds to the decomposition of the soundfield into a numerable set of compact acoustic beams in the ray space domain. Thanks to this representation, acoustic objects (i.e., sources and reflectors) become linear features, thus enabling pattern analysis techniques also with the aim of Deep Learning methods. At the same time, representations based on Spherical Harmonic (SH) decompositions have gained popularity in the research community, as they enable the definition of Higher-Order Microphones (HOM) and Loudspeakers (HOL), which are compact clusters of transducers that are able to capture SH components up to a given order. This has the advantage of relaxing constraints on spatial sampling, at the cost of a higher computational cost and reduced flexibility.
Is there a middle ground where the flexibility and efficiency of RST synergistically combine with the reduced invasivity of HO processing?

The objective of this research is to combine the advantages of the RST with the Spherical Harmonics domain in a unified framework for the soundfield analysis and processing. Furthermore, to enable an acoustic interaction with the environment, information about the vibrational characterization of sources is needed. In this sense, Near-field Acoustic Holography (NAH) enables an accurate prediction of the vibrational field on objects in a fully contactless way. Therefore, this research aims to explore new Deep Learning based solutions for NAH, inspecting the link between the information gathered by Neural Networks and the beam distribution which models the environment.