Effect of control samples and listener attributes on speech naturalness ratings of people who stutter

Speech restructuring treatment can effectively reduce stuttering but the resultant speech may sound unnatural. Martin et al. (1984) speech naturalness scale is widely used by clinicians and researchers, yet little is known about whether including normally fluent speech samples alters the judgement of the naturalness of speech samples of people who stutter, and whether attributes of listeners - specifically training and sex - influence ratings.

In this study 20 untrained listeners (male and female) and 19 speech language pathology students (female only) rated either the naturalness of 21 speech samples from adults who stutter obtained post-treatment, or the same 21 post-treatment samples randomly mixed with samples of 21 samples from normally fluent speakers matched for age and sex. The independent variables were sample composition (addition of fluent controls) and listener training. The dependent variable was listener naturalness rating.

A two-factor ANOVA with listener training and sample composition as independent variables and naturalness ratings as the dependent variable was performed. Untrained listeners rated samples as significantly less natural than trained listeners. The addition of control samples did not significantly impact scores assigned to post-treatment samples. A comparison of male and female listeners was completed using the Mann Whitney U Test. A significant group difference was observed with female listeners rating the samples more leniently (more natural) than male listeners.

Based on this preliminary research, the addition of controls does not appear necessary in evaluating speech naturalness, however the composition of the listener group may affect results.