With the advent of smartphones and high-speed internet, the massive amount of data is all around. Almost every aspect of life is now being recorded as data: health monitoring devices, various entertainment mobile apps, banking data, social media networks, marketing advertisements, and the list goes on. The complexity of such networks and interactions means that the data now resides on irregular and complex structures that do not lend themselves to standard tools. Graphs offer the ability to model such data and complex interactions among them. For example, users on Twitter can be modelled as nodes while their friend connections can be modelled as edges. [ref]

The sound source localization problem can be considered as a sparse recovery problem, where sparsity is enforced in terms of few active sources among all candidate direction of arrivals (DOAS) from predefined angular grid. Compressive sensing (CS) is a paradigm to address such sparse inversion problems with few noisy linear measurements by enforcing sparsity on unknown signal. We formulate sparse signal models and develope CS based algorithms focusing along sparse Bayesian learning based algorithms.

The localization of acoustic source is often considered as a direction-finding problem, where the aim is to find the direction of arrivals (DOAs) of the source. Localization provides estimates of source locations (DOAs) at specific time instants. Along with localization, it is typically desired that the sources are tracked over time. This is important because sources move around spatially and hence connecting the localization results over time provides a complete trajectory of the source. Considering applications like smart cars, hearing aids, smart homes, drones etc., development and benchmarking of the localization and tracking algorithms on these application scenarios are of significant importance and an active area of research. All the smart devices are embedded with microphone array to record audio signals which are often affected by reflection, reverberation, background noise, and multi-path effects which makes the localization challenging. We use compressives sensing based localization algorithms and recently deep learning based methods to solve acoustic source localization and tracking problems.