On semi-blind estimation of echo paths during double-talk based on nonstationarity

The estimation of a filter that determines an echo path is a difficult problem when double-talk is present. We use a Blind Source Separation model based on signals´ nonstationarity with a partially known mixing matrix to estimate the filter during the double talk. A second-order approximation of a log-likelihood function is used to derive a quadratic criterion. Then, we propose two methods to estimate the filter through minimizing the criterion. The first method works in a batch-offline regime, while the second method is designed to estimate the filter adaptively. The methods are verified by experiments and compared with standard least-squares-based approaches.