Speech enhancement using a frame adaptive gain function for Wiener filtering

Many existing speech enhancement techniques, especially Wiener filtering, suffer from introducing annoying musical noise and speech distortion in low SNR due to their rigid gain functions. In this paper we propose a modification to the parametric Wiener filter that emphasizes the spectral contributions in spectral regions which are important for intelligibility. This is done by defining an adaptive parameter that is a function of the pitch. Objective measures and statistical tests are used to assess subjective speech quality and intelligibility. The results indicate that the proposed algorithm results in speech intelligibility improvement and in musical noise reduction, as compared to the parametric Wiener filter.