Speech-Interference Criteria

Abstract

An attempt has been made to show the similarities between three ostensibly different methods of rating noises for speech interference. The three basic methods are: sound-level meter (SLM) readings, using various frequency-weighting networks; noise criteria contours, where spectral peaks of noise become tangent to one of a family of rating curves; and dB-averaging methods, the articulation index (AI) being the most sophisticated method and the speech-interference level (SIL) being the simplest to use. A speech-interference (SI) curve has been evolved that, when used as a frequency-weighting network in a SLM, or as a noise-rating curve, or as a curve-fitting method of arriving at an SIL, greatly reduces the spread of scores among the three measurement methods when rating the speech-interfering properties of noise. Based on a literature review of speech intelligibility in noise, a family of SI contours was evolved to encompass greater ranges of noise levels (and intelligibility scores). These contours are compared to a theoretical treatment of the AI in idealized noises. Many similarities are noted. One potential use of the SI curve (as a frequency-weighting network in a SLM) is discussed: namely, a method of locating potentially objectionable noise environments from SLM readings alone.