A robust detector is proposed for detection of the consistency of two speech signals. The detection is based on comparing the pitch periods of the speech signals. The distribution of the noise is assumed to be a mixture of Laplacian distribution, giving a sharp peak around the true value of the signal of interest and uniform distribution modelling the contributions of completely unknown noise. This type of noise appears in problems related to estimated pitch periods. We derive a robust detector for this noise model and analyse its performance.