Estimating the Intelligibility of Continuous Discourse
Charles Speaks · The Journal of the Acoustical Society of America · 1969
Performance on a synthetic sentence identification (SSI) task was compared with data from Hawkins and Stevens (1950) for continuous discourse. First, the relation of speech level to noise level for an SSI performance criterion of 50% correct response was determined. This function then was compared with similar functions representing the thresholds of intelligibility (TI) and detectability (TD) for continuous discourse. The SSI function lies approximately midway between TI and TD for continuous discourse. When the Hawkins and Stevens definitions of threshold were applied to SSI data, TI for continuous discourse corresponded to approximately 99% and TD to 5% performance for SSI. Masked thresholds of intelligibility for continuous discourse can be estimated accurately by determining the lowest level at which a listener can perform the sentence task without error. The close correspondence suggests that when intelligibility of continuous discourse is difficult to quantify, reasonable estimates can be gained from other test materials if the gain functions are parallel to those of continuous discourse for a given experimental condition. [Supported by Graduate School, University of Minnesota, and by the Vocational Rehabilitation Administration.]