A RATING SCALE TECHNIQUE FOR THE MEASUREMENT OF SPEAKER INTELLIGIBILITY

An evaluation was made of the method of measuring speaker intelligibility by listener ratings of voice samples on an equal-appearing intervals scale. Twenty-four speakers and seven panels of listeners, with a minimum of 20 persons in each panel, were involved in the experiment. Recordings were made of each speaker reading multiple-choice intelligibility test word lists and prose material. Ten-second voice samples were prepared from the prose reading. The multiple-choice test material was played for listener panels to provide for each speaker a percent intelligibility score. The ten- second voice samples were played for listening panels under various listening conditions to provide for each speaker a scale value intelligibility score. These listening conditions were that of hearing the voice signal in quiet and at the SN ratios of 5 db, 0 db, and -5 db. Correlation coefficients were determined between multiple-choice and scale value scores to provide an estimate of the validity of the rating method. An analysis of variance was used to test the significance of the differences among the mean scale values with respect to the different listening conditions. Moderately high positive correlations between multiple-choice and scale value intelligibility scores suggest that the rating scale method provides a fairly good estimate of speaker intelligibility. Q-values, which provide an index of reliability, were within reasonable limits.