Development of normed speech-in-noise sentence lists for use with children

Listeners with equivalent pure tone audiograms can vary considerably in their ability to perceive and understand speech-in-noise. In addition to perceptual abilities, this variation may reflect differences in language or cognitive skills which support processing of the noise-degraded signal. The R-SPIN sentences (Bilger, et al., 1984) were developed to capture these differences and help researchers distinguish between perceptual and linguistic contributions to listening-in-noise. This is done by varying the extent to which the final word can be predicted from the preceding context. We recently developed a similar set of sentences (BESST-UK) which were additionally carefully matched for prosody, length and overall structure (BESST-UK, Heinrich, et al., 2014; Barry, et al., 2014). Here, we describe further work aimed at developing normed lists of sentences drawn from the BESST-UK database for use with children.

First, 250 children, subdivided into three age bands (4-6, 7-8, and 9-12 years), listened to and repeated sentences presented in 12-talker babble at up to three different signal-to-noise ratios (SNR). The aim was to identify the SNR resulting in scores in the range of 70 to 90 percent correct (unpredictable versus predictable sentences respectively) for each age band. As has been shown in previous studies, young children needed a more favorable SNR (+4 dB) to achieve the same level of performance as the oldest group of children (-1 dB).

In the next step of the project (in progress), the data at the age-specific preferred SNRs will be modelled using Rasch analysis. This is a form of latent trait modelling which is applied in measurement development. It estimates sensitivity of individual items to levels of ‘difficulty’ in a particular latent trait. Rasch analysis has not been applied to speech perception data. However, we predict that sentences drawing on cognitive / language abilities (i.e. predictable sentences) will be more amenable to modelling, than sentences where perceptual abilities are being tested (i.e., unpredictable sentences). We hope to use the results of the Rasch analysis to develop multiple sentence sets which offer equivalent measures of speech perception abilities.

Barry, et al. (2014). Sensitivity of the British English Sentence Set Test (BESST) to differences in children’s listening abilities. BSA Conference.
Bilger, et al. (1984). Standardization of a Test of Speech-Perception in Noise. JSLHR, 27(1), 32-48.
Heinrich, et al. (2014). Assessing the effects of semantic context and background noise for speech perception with a new British English speech test. BSA Conference, Keele.