Method and apparatus for speech processing are disclosed. A first unvoicing parameter for a first frame of a speech signal is determined, and furthered smoothed based on a second unvoicing parameter for a second frame prior to the first frame. A difference between the first unvoicing parameter and the smoothed unvoicing parameter for the first subframe is computed and a unvoiced/voiced classification of the first frame is determined using the computed difference as a decision parameter. Further processing, such as Bandwidth extension (BWE) is performed on based on the classification of the first frame.