Reinforced spectral subtraction method to enhance speech signal

摘要

In this paper, we propose a speech enhancement technique. It is based on spectral subtraction method, which is one of the major techniques of speech enhancement. However, the enhanced output speech signal of the spectral subtraction method is corrupted by "musical noise". The musical noise is an offensive noise for human listening. To reduce the musical noise, we adopt an iterative algorithm. The iterative algorithm is derived from the same idea as Wiener filtering for speech enhancement. We use the output signal of the spectral subtraction method as the input signal again. This process is iterated a few times. At every time when we iterate the spectralsubtraction method, we estimate the noise signal, in order to obtain a peculiarity of musical noise sad subtract it Therefore we can reduce more musical noise for iterations.