Musical noise suppression using isolated factor of power spectrum

Most speech enhancement algorithms based on weighting gains may generate annoying musical noise in a certain degree. In this paper, we propose a correction factor for modifying the weighting gains, which was based on the isolated factor. The isolated factor indicates the isolated degree of the local power to its neighbouring bins. The smaller isolated factor indicates the greater degree of the local power relative to its neighbouring bins. The proposed algorithm could distinguish musical noise correctly and suppress it efficiently, and could be applied to all the kinds of algorithms that basing on the weighing spectral gains. PESQ test and SegSNR improvement experiment support the conclusion.