Abstract
A coherence function based dual-microphone source separation method in the adverse noisy environments is presented in this paper. The method is based on the coherence between the clean speech and noise signal for the noise reduction. The technique is applied to closely spaced microphones array and assumed that noise signals are captured by microphones are highly correlated. The presented technique is straightforward to realize and does not need any noise statistics estimation. The technique is evaluated in terms of the speech intelligibility, perceptual quality, segmental SNR (SNRSeg), speech distortion and residual noise objective measures. The experimental results indicate improved gains in perceptual quality, and speech intelligibility comparative to the noisy speech. Moreover, the presented technique yielded substantial improvements in terms of the SNRSeg, speech distortion and residual noise, respectively in low SNR (-5 and -10 dB) noisy environments than noisy speech. The time-varying spectral and time-domain waveform analysis indicated a low residual noise and less speech distortion in the spectrums of speech processed by the coherence based source separation technique.