Child Speech Genre Classification and their Potential Application in Monitoring Technologies for Safety Purposes

Abstract

In an ever more demanding world, where parents and educators alike struggle to provide attention to young children, the advance of technologies to analyze child speech and monitor its development, along with that of children’s themselves, is paramount. This study aims to distinguish two well studied and documented genres in child speech, conversation and narrative. We propose here a straightforward, yet effective method which relies on the comparison of the most common bi-grams, typical two-word sequences. This classifying method is applied to four different corpora, two for Spanish and two for English, with one data set for each of our two targeted genres. Our method uses a semi-automatic extraction technique to obtain bi-grams and successfully compares their distinctive distribution using the statistical test of chi-square. The potential applications of a semi-automated classification of these genres are varied, from child development diagnosis and improvement, to monitoring of children’s safety and well-being.