Modified Spectral Subtraction Based Speech Enhancement

The one-microphone speech enhancement, (SE) algorithm as a modification of the extended spectral subtraction (SS) method (ESS) [1] is presented in this paper. The algorithm can be used for the reduction of additive stationary and quasi-stationary colored broad-band noise in noisy speech in hands-free communication terminals as well as in a number of other applications. The achieved noise reduction of the algorithm is comparable to other onemicrophone methods. However, the proposed algorithm produces a lower level of ”musical noise”, less than that of the known ones. The proposed algorithm does not require a voice activity detector (VAD), and is very simple (it should be implementable in the bounds of 10 MIPS computing resource in a DSP platform, similar to TMS320C50x).