Detecting double-talk and path changes in echo cancellers

To detect double-talk and path changes in an echo canceller, a correlation-based matrix of signals is generated. Analysis of the matrix enables identification of double-talk and path changes. The matrix may be generated using zero-lag auto- and cross-correlations. Double-talk and path changes are inferred when a determinant of the matrix passes a threshold. Alternatively, eigendecompositions or single value decompositions of the matrix may be used. The elements of the matrix may be generated in the time or in the frequency domain. The canceller may employ an adaptive filter implementing a normalised-LMS algorithm.