Noise reduction by joint maximum a posteriori spectral amplitude and phase estimation with super-Gaussian speech modelling

For acoustical background noise reduction a computationally efficient joint MAP estimator with a super-Gaussian speech model is presented. Compared to a recently introduced MAP estimator the new joint MAP estimator allows an optimal adjustment of the underlying statistical model to the real PDF of the speech spectral amplitude. The computationally efficient estimator outperforms the Ephraim-Malah estimator and the recently proposed MAP estimator in a single microphone noise reduction framework due to the more accurate statistical model.