Single–Channel Speech Dereverberation based on Spectral Subtraction

Speech signals recorded with a distant microphone usually contain reverberation, which degrades the fidelity and intelligibility of speech in devices such as ’hands–free’ conference telephones, automatic speech recognition and hearing aids. One important effect of reverberation on speech is overlap–masking, i.e. the energy of the previous phonemes is smeared over time, and overlaps following phonemes. In [1] a single–channel speech dereverberation method based on Spectral Subtraction was introduced to reduce this effect. The described method estimates the power spectrum of the reverberation based on a statistical model of late reverberation. This model depends on one parameter, the reverberation time. However, the reverberation time is frequency dependent due to frequency dependent reflection coefficients of walls and other objects and the frequency dependent absorption coefficient of air. In this paper, we have taken this dependency into account and studied the effect on reverberation reduction and distortion. The algorithm is tested using synthetically reverberated signals. The performances for different room impulse responses with reverberation times ranging from approximately 200 to 1200 ms show significant reverberation reduction with little signal distortion. Keywords— dereverberation, spectral subtraction, speech enhancement.