Assessment of Statistical Array Geometries for Covert Surveillance

The proposed work seeks to provide criminal justice and law enforcement agencies with enhanced abilities to covertly record and listen to remote conversations in noisy areas, which include interfering conversations. The ultimate target for the technology involves the combination of microphones mounted around an area of interest or on moving platforms (i.e. undercover agents, trained animals, or small robotic devices) with parallel/distributed computing to provide near-real time processing for identifying speech sources and extracting intelligible speech for eavesdropping or recording. The goals of this work extend the current state-of-the-art to analyses of distributed microphones systems that are not in a regular geometry with subjective assessments of intelligibility enhancement. Assessment will determine the impact of distributed microphone systems on improving intelligibility of conversations in noisy rooms, which include interfering conversations, reverberation, and non-speech noises. Intelligibility enhancement from with regular and random microphone geometries will be examined through Monte Carlo simulations and physical experiments with subjective as well as numerical evaluations. Relationships between intelligibility, levels of interference/noise, and microphone distribution geometries will be described. In particular the conditions for achieving acceptable intelligibility in a variety of noise environments will be presented based on physical experiments with subjective listening tests.