Improving Statistical Model-Based Speech Enhancement with Deep Neural Networks

Systems and methods improving the performance of statistical model-based single-channel speech enhancement systems using a deep neural network (DNN) are disclosed. Embodiments include a DNN-trained system to predict speech presence in the input signal, and this information can be used to create frameworks for tracking noise and conducting a priori signal to-noise ratio estimation. Example frameworks provide increased flexibility for various aspects of system design, such as gain estimation. Examples include training a DNN to detect speech in the presence of both noise and reverberation, enabling joint suppression of additive noise and reverberation. Example frameworks provide significant improvements in objective speech quality metrics relative to baseline systems.