Quantifying continuous child speech for automated detection of speech impairment

Abstract

The SpeechMark® Automated Syllabic Cluster detection system was tested as a novel approach for analysis of continuous speech samples recorded from 4-year-old children classified as typically developing (TD, N = 44, M = 4.32 years, SD = 0.64) and with speech compromise (SC, N = 16, M = 4.14 years, SD = 0.66). The speakers were recruited in the Midwest and Southern regions of the United States. To test if the TD group produced higher syllabic clusters compared to the SC, we fit a generalized linear mixed effects model. The model adjusted for the potential influence of age and dialect in contributing to the group differences by including them as covariates. Results were interpreted using incidence rate ratios (IRR). Results showed that the IRR was dependent on age indicated by the significant interaction term group*age (p-value = 0.003). The results also showed that there was no difference between the two dialect groups. Results from linear mixed effects models showed that the speech rate was higher among speakers in SC group given all other factors held constant (effect = 0.9, p-value = 0.055). These findings are promising as we aim to automate the analysis of continuous speech samples of young children.