Single and multichannel enhancement of distant speech using characteristics of speech production

Speech collected at a distance from a speaker is degraded due to background noise, reverberation and other audio/speech signals. Normally enhancement of the degraded signal focuses on determining the characteristics of degradation and auditory perception. There are many characteristics of speech production which contribute to robustness against degradation caused by distance between the speaker and the pickup microphone. One such characteristic is the sequence of impulse-like excitation of the vocal tract system during production of speech. This paper proposes single channel (mic) and multichannel (mics) enhancement methods for processing the distant speech. The single channel enhancement method exploits the pitch periodicity and the epoch locations to reduce the effect of additive background noise. A multichannel enhancement method is proposed not only to reduce the effect of background noise, but also to reduce the effects due to reverberation and other weak interfering signals. The multichannel method exploits the invariance property of the sequence of impulses due to excitation in the direct component of the sound at the two microphone locations. The results of enhancement are available at http : //speech.iiit.ac.in/index.php/downloads.html/hscma2011/.