A T Location-scale-based MMSE estimator under a combination of stochastic and deterministic speech model

In recent years many speech enhancement methods are based on stochastic models. In contrast with most recently models, we do not consider that DFT coefficients may be modeled just by complex Gaussian, Laplacian or Gamma density and stochastic character. We model the DFT coefficients by T location-scale distribution. Then we derive a new MMSE estimator which relies on stochastic-deterministic model [1]. At last, we compare it with previous stochastic-deterministic models and demonstrate our model performance improvement.