Speech enhancement using a joint map estimator with Gaussian mixture model for (non-)stationary noise