Learning-Based Source Separation Methodologies Applicable to the Multiple Target Problem

"Coprime proposes to investigate the applicability of learning-based source separation methodologies to the problem of multiple targets in a complex acoustic environment. The Phase I effort will focus on identifying viable learning-based source separationalgorithms specifically tuned to the multiple combat vehicle scenario. A prototype architecture will be developed and numerical software rapidly developed in order to test the performance of the learning-based algorithms developed on multiple combatvehicle data sets. Coprime proprietary Vehicle Signal Analysis Environment and Vehicle Monitor Simulation Environment will be employed to assist in evaluating the performance of the learning-based source separation algorithms developed under this Phase I.Anticipated benefits include robust, low cost, software-based, modular algorithms that can be rapidly integrated into a wide variety of passive, acoustic monitors used in sophisticated surveillance systems."