Passing the Strongly Voiced Components of Noisy Speech

Abstract

RUNNING speech can be processed electronically so as to leave only the strongly voiced speech sounds, with silent gaps at other points in the speech continuum. This “gating” is achieved by detecting energy in a bandwidth corresponding to the presence of strongly voiced speech sounds; that is, 400–800 Hz. The detector controls an amplitude-limiter which blocks all weakly voiced components while passing the strongly voiced components with an effectively unlimited bandwidth. Such gated speech sounds stultified and is markedly less intelligible than the original. Cherry and Wiley have demonstrated, however, that the addition of white noise in the silent gaps of the gated speech can increase its intelligibility^1,2, and evidence has been adduced that this is because of a restoration of a rhythmic pattern in the speech.