Uniform Speech-Peak Clipping in a Uniform Signal-to-Noise Spectrum Ratio

Abstract

A graphical function W(r, c) has been determined experimentally, in which W is word articulation, r is the relative level of unclipped speech and the noise, and c is the amount of uniform, symmetrical, speech-peak clipping. Preemphasis of the speech signal gave an approximately uniform speech spectrum prior to clipping. Uniform, random noise was mixed with the clipped speech before post-equalization, making the final noise spectrum similar in shape to the speech spectrum. The real-ear response of the earphones was compensated electrically to yield a uniform orthotelephonic response for the communication system, in the frequency range contributing significantly to articulation index. For constant clipping cn the function W(r, cn) approaches W(r, 0) as a limit for sufficiently large values of r. For c < r + 5, W ≑ W(r, 0). For the case of no clipping W(r) when transformed to W(A), A being articulation index, resembles closely the curve by Pollack.