R & D Activities

During the last few years, SPL members have been active in original research in 6 major axes:

In particular, SPL research activities are structured around several interwoven themes:

STATISTICAL SIGNAL PROCESSING

The Statistical Signal Processing activity focuses on the development of advanced algorithms and systems for multimedia content manipulation and delivery over broadband wireless networks. Research topics of interest include time series analysis, compressed sensing, image and signal compression, and statistical communication theory. The unifying theme of this activity is the application of statistical theory to characterise the operational environment and to investigate, develop, integrate, and validate novel techniques for information processing and transmission. In particular, related R&D activities focus on:

Multiresolution signal modeling, denoising, compression, and content-based information retrieval
Research activities at SPL explore the statistical properties of linear time-frequency analysis methods, which include the short-time Fourier transform, wavelets and the discrete cosine transform (DCT). SPL work demonstrates that subband decompositions of actual signals have significantly non-Gaussian statistics that are best described by families of heavy-tailed distributions, such as the alpha-stable. Modelling information is used to design optimal processors that exploit the signal statistics by performing non-linear operations on the data. The adopted approach is unique in that it relates the optimal non-linearity to the degree of non-Gaussianity of the data. Such a methodology is applied to a wide range of problems, including medical and SAR image denoising and autofocusing, multiresolution data compression, blind watermark detection, and texture image retrieval.
Collaborative detection, classification, and tracking
Research and development activities concentrate on self-organisation of heterogeneous sensors in unstructured and uncertain environments, and the fusion of the information for intelligent learning and decision making. The key issue of concern is the fusion of sensor data when the relationships among the signals sensed by different sensors, the character of those signals and the environment in which they propagate are uncertain, variable, or simply unknown. SPL develops distributed pattern matching algorithms that are robust to these uncertainties, and that can learn from the observed data and then adapt based on what is learned.
Distributed signal acquisition and representation
Coherent processing issues for real wireless sensor network systems are investigated to enable their operation in a wide range of adverse environmental conditions. Beamforming methods have been developed to collect the power of the dominant sources while also providing high rejection of interference and noise; the objective is to design optimal maximum-likelihood (ML) methods and suboptimal, computationally efficient distributed techniques using sub-arrays yielding cross-bearing information, which will be employed to perform accurate source localisation. The ultimate goal is to allow the network to self organise and dynamically configure the needed sensor nodes to perform complex beamforming operations.

AUDIO SIGNAL PROCESSING

An important area of signal processing includes the investigation of audio signals, given their specific harmonic structure and the importance of these signals in a variety of people’s everyday activities. SPL has significant related expertise having derived important research results and algorithms in diverse applications, such as the acquisition, coding, transformation, separation, retrieval, enhancement, and 3D rendering of audio signals.

Spatial sound acquisition and rendering methods for immersive environments
Immersive audio systems have been making strides in such applications as telepresence, augmented and virtual reality, entertainment, distance learning, and sound editing for television and film. SPL performs research on signal processing issues that pertain to the acquisition and subsequent rendering and transmission of 3D sound fields. On the acquisition side, advanced statistical methods have been developed for achieving acoustic arrays in audio applications, by addressing two major aspects of spatial filtering, namely localisation of a signal of interest and adaptation of the spatial response of an array of sensors so as to achieve steering in a given direction. On the rendering side, surround sound and 3D audio methods have been developed that generate virtual sound sources around the listener.
Audio signals analysis and modelling
The analysis and modelling of audio signals finds important applications in a variety of fields, such as audio compression, sound source separation and enhancement, music analysis and retrieval, and so forth. SPL has developed algorithms which analyze and exploit properties of music signals, such as the harmonic structure of sound signals, the temporal and spectral variations of music signals with respect to time, and spatial attributes in the spectral information. Applications of interest include multichannel audio compression, separation of non-linear mixtures of sound sources, and tempo estimation and onset detection for the analysis and classification of music signals
APPLICATION AREA: Sensor Networks for Immersive Environments
A focused application area in SPL lies in the intersection of wireless sensor networks and immersive audio environments. We would like to enable “immersive presence” for the user at any event where sound is of interest. This includes concert-hall performances; outdoor concerts performed in large stadiums; wildlife preserves and refuges, studying the everyday activities of wild animals; and underwater regions, recording the sounds of marine mammals. In the case of large venues which are possibly outdoors or even underwater, and for recording times of days or even months, the traditional practice of deploying an expensive high-quality recording system which cannot operate autonomously becomes impossible. The capture, processing, coding and transmission of the audio content through multiple sensors, as well as the reconstruction of the captured audio signals so that immersive presence can be facilitated in real time to any listener, are central application goals of SPL researchers.

SPEECH SIGNAL PROCESSING

SLP performs research in the area of speech signal modelling and has proposed or adapted several mathematical and statistical models for representing the time-frequency properties of speech signals, and for analyzing speaker-dependent information. Statistical transformations of speech spectral features have also been investigated. SPL has applied the derived models in several applications of interest, including voice conversion, speech enhancement, speaker identification, speech emotion recognition, speech coding, and voice pathology.