Intelligibility Threshold Level (ITL) Ratings of Some Current Digital Voice Communications Processors

摘要

Intelligibility scores for digital voice communications processors have been found to be typically characterized by highly significiant differences among speakers, as well as highly significant differences among scores for the various phonetic features. Distributions of intelligibility scores for speech processors are not normally distributed, but highly skewed. A negative binomial probability distribution was found to give good agreement with empirical data distributions of speech intelligibility scores. A new performance rating for voice communications devices, termed an intelligibility threshold level (ITL), was conceived as a means of taking these findings into consideration in establishing a measure of intelligibility performance that is an estimate of an intelligibility value that the majority (rather than the simple average) of intelligibility scores for a voice processor will equal or exceed, at a specified confidence level established in relation to the sample size used in obtaining the rating. It is proposed that an ITL rating is a more meaningful assessment of the degree of risk involved in misunderstanding voice messages or causing time to be lost in requiring messages to be repeated. It was shown that ITL's can be determined by two alternative methods: by rank-ordering the intelligibility scores for a processor and constructing the cumulative distribution of data and its confidence band, or by using a negative binomial probability model for the data distribution.