A S-transform based spectrum enhancement method for complex noise environment

Efficient spectrum enhancement plays an important role in speech recognition. The present system can work well in quiet environment or white Gaussian noise, but degrades quickly in mixed noise environment. To solve this problem, this paper presents a new spectrum enhancement method based on generalized S-transform. Firstly, the influence of α-stable noise is reduced by the robust S-transform based on L-DFT. Then, to resolve the interference of the intensified Gaussian noise, a varying-tendency window is introduced into the S-transform. By this way, we can get the better spectrum enhancement under the mixed noise environment. The effectiveness of the presented method is verified by simulation.