ISCA Archive Interspeech 2023
ISCA Archive Interspeech 2023

Using Random Forests to classify language as a function of syllable timing in two groups: children with cochlear implants and with normal hearing

Mark Gibson

We trained a series of Random Forest models in a supervised learning environment on different temporal parameters related to syllable structure: voice onset time (VOT), vowel duration following simplex and complex onsets, and lateral duration in word initial (/lV) position and as the second consonant in a C1C2 cluster (where C means consonant). Capitalizing on previous work we trained the models on data from monolingual Spanish- and English-speaking adults. We asked whether the timing productions used by bilingual children with normal hearing (NH) and children with cochlear implants (CI) can be classified as pertaining to the same timing system (i.e. language), or whether the children are applying the same basic timing plan to two different languages. We also asked whether there were differences between the CI and NH groups. Our results indicate that the children from both groups produce qualitatively distinct timing plans for each language with no interference from the other language.