Relation between intelligibility scores for four test methods and three types of speech distortion.

Abstract

Several intelligibility tests are currently being used to evaluate speech-communication systems. Individuals engaged in the development of such systems would often like to compare the performance of a system evaluated with one test with the performance of another system that has been evaluated with a different test. A study has been conducted to examine and quantify the relation between scores obtained with four intelligibility tests (the Harvard PB-Word Test, the Fairbanks Rhyme Test, the Modified Rhyme Test, and the Harvard Test Sentences) for various levels of three types of speech distortion (additive speech-shaped noise, peak clipping, and vocoderization). Recordings by two adult male speakers of the different test materials were processed by the three distortion systems, and output recordings were administered to a group of listeners. The results of the study demonstrate that the relation between the various test scores is not unique but depends considerably on the type of speech distortion employed.