Generalized Learning of Dysarthric Speech Between Male and Female Talkers

Abstract

Purpose Perceptual training is a listener-targeted means for improving intelligibility of dysarthric speech. Recent work has shown that training with one talker generalizes to a novel talker of the same sex and that the magnitude of benefit is maximized when the talkers are perceptually similar. The current study expands previous findings by investigating whether perceptual training effects generalize between talkers of different sex. Method Forty new listeners were recruited for this study and completed a pretest, familiarization, and posttest perceptual training paradigm. Historical data collected using the same three-phase protocol were included in the data analysis. All listeners were exposed to the same talker with dysarthria during the pretest and posttest phases. For the familiarization phase, listeners were exposed to one of four talkers with dysarthria, differing in sex and level of perceptual similarity to the test talker or a control talker. During the testing phases, listener transcribed phrases produced by the test talker with dysarthria. Listener transcriptions were then used to calculate a percent words correct intelligibility score. Results Multiple linear regression analysis revealed that intelligibility at posttest was not predicted by sex of the training talker. Consistent with earlier work, the magnitude of intelligibility gain was greater when the familiarization and test talkers were perceptually similar. Additional analyses revealed greater between-listeners variability in the dissimilar conditions as compared to the similar conditions. Conclusions Learning as a result of perceptual training with one talker with dysarthria generalized to another talker regardless of sex. In addition, listeners trained with perceptually similar talkers had greater and more consistent intelligibility improvement. Together, these results add to previous evidence demonstrating that learning generalizes to novel talkers with dysarthria and that perceptual training is suitable for many listeners.

Figure 1.

Pretest and posttest percent words correct (PWC) scores by talker condition. The mean pretest and posttest PWC scores for each condition are depicted above. Data from the control, male similar, and male dissimilar conditions came from the original study (Borrie et al., 2017a). Data for the female similar and dissimilar conditions were collected for this current study. Notably, all listeners trained with dysarthric talkers had greater PWC scores on posttest compared to listeners trained with a control talker. Listeners trained with similar talkers had the greatest PWC improvement compared to the listeners trained with dissimilar talkers.

Figure 2.

Distribution of improvement scores paneled by level of similarity. The above figure shows the intelligibility improvement scores for listeners from both the original study (Borrie et al., 2017a) and this current investigation. The figures show distributions of intelligibility improvement across talker sex, with improvement scores from listeners trained with the control talker (top), the dissimilar talkers (middle), and the similar talkers (bottom). The listeners trained with the dissimilar talkers showed greater variability in intelligibility improvement compared to the listeners trained with the similar talkers, evidenced by a flatter distribution and by coefficient of variation analysis results (see Table 3).