An automated classifier for periods of sleep and target-child-directed speech from LENA recordings
Some theories of language development propose that children learn more effectively when exposed to speech that is directed to them (target child directed speech, tCDS) than when exposed to speech that is directed to others (other-directed speech, ODS). During naturalistic daylong recordings, it is useful to identify periods of tCDS and ODS, as well as periods when the child is awake and able to make use of that speech. To do so, researchers typically rely on the laborious work of human listeners who consider numerous features when making judgments. In this paper, we detail our efforts to automate these pro-cesses. We analyzed over 1,000 hours of audio from daylong recordings of 153 English- and Spanish-speaking families in the U.S. with 17- to 28-month-old children that had been previously coded by hu-man listeners for periods of sleep, tCDS, and ODS. We first explored patterns of features that character-ized periods of sleep, tCDS, and ODS. Then, we evaluated two classifiers that were trained using auto-mated measures generated from LENATM, including frequency (AWC, CTC, CVC) and duration (mean-ingful speech, distant speech, TV, noise, silence) measures. Results revealed high sensitivity and speci-ficity in our sleep classifier, and moderate sensitivity and specificity in our tCDS/ODS classifier. Moreo-ver, model-derived predictions replicated previously-published findings showing significant and posi-tive links between tCDS, but not ODS, and children’s later vocabularies (Weisleder & Fernald, 2013). This work offers promising tools for streamlining work with daylong recordings, facilitating research that aims to better understand how children learn from everyday speech environments.
Keywords: child-directed speech, other-directed speech, LENA, daylong recordings, automated classifier