Single channel speech enhancement using a new binary mask in power spectral domain

The paper proposed by kim et al uses a new binary mask approach to improve speech intelligibility based on noise distortion constraints. In our paper we have proposed to apply the same technique in power spectral domain to study its effects on the overall speech enhancement aspects. Subjective and objective tests were conducted to evaluate the performance of the enhanced speech for various types of noises. The experimental results indicated the improvement in segmental signal-to-noise ratio (SSNR) at 0 and -5 dB input signal-to-noise (SNR) values for helicopter noise. The results are identical for other types of noises such as babble and random noise in both the domains. It is also observed that there is a boost in terms of intelligibility even at low SNR levels (i.e. -5 dB) in the power spectral domain for helicopter noise. We have compared the results with that implemented in magnitude spectral domain.