Unknown signal detection by one-class detector based on Gaussian copula

One-class detector is an option to deal with the problem of detecting an unknown signal in a background noise, as it is only necessary to know the noise distribution. Thus a Gaussian copula is proposed to capture the dependence among the noise samples, meanwhile the marginals can be estimated using well-known methods. We show that classical energy detectors are particular cases of the proposed one-class detector, when Gaussian noise distribution is assumed, but are inappropriate in other cases. Experiments combining simulated noise and real acoustic events have confirmed the superiority of the proposed detectors when noise is non-Gaussian. An interpretation of the methods in terms of the Edgeworth expansion is also included.