Talker variance and phonetic feature variance in diagnostic intelligibility scores for digital voice communications processors

Multiple-speaker tests of digital voice communications processors typically result in intelligibility scores deviating significantly from normal distributions, and characterized by significant differences between talker mean scores, as well as between mean scores for phonetic features. Rank-ordering of populations of diagnostic scores for a processor, and preparation of cumulative distribution functions (CDF´s) facilitates both formal and informal assessments of intelligibility performance. Findings suggest that it would be appropriate to re-assess questions of the appropriate number of talkers and the basis for their selection for formal intelligibility testing.