Intelligibility improvements obtained by an enhancement method applied to speech corrupted by noise and reverberation

Abstract

A series of psychoacoustic experiments are described which attempt to assess the capability of a Multi-Microphone Sub-band Adaptive Signal (MMSBA) processing scheme, for improving the intelligibility of speech corrupted with noise and reverberation. The processing scheme applies the Least Mean Squares (LMS) adaptive algorithm in frequency delimited sub-bands to process speech signals from simulated and real room acoustic environments with various realistic signal to noise ratios (SNR). The processing scheme aims to take advantage of binaural input channels to perform noise cancellation. The two wide-band signals are split into linear or cochlear distributed sub-bands, then processed according to their sub-band signal characteristics. The results of a series of intelligibility tests are presented in which speech and noise data, generated in simulated and real room conditions, was presented to human volunteer subjects at various SNRs, sub-band distributions and sub-band spacings. The results from both simulated and real room acoustical environments show that the MMSBA processing scheme significantly improves both SNR and intelligibility.