ISCA Archive SpeechProsody 2024
ISCA Archive SpeechProsody 2024

Real-Time Relations Between Prosodic Features of Infant-Directed Speech and Infant Attention at 3 Months

Yannan Hu, Mark Hasegawa-Johnson, Nancy McElwain

Infant-directed speech (IDS) guides infants’ attention during caregiver-infant interactions. Prior studies have demonstrated infant preference for IDS over adult-directed speech and linked fundamental frequency (F0) and its variability to infant attention. However, we know little about how various prosodic features predict real-time infant attentional responses. In this study, we assessed the IDS prosodic features that best predicted infant attentional engagement during mother-infant play. Seventy-nine infants and their English-speaking mothers were video-recorded during an 8-minute play session at 3 months. From each IDS utterance, we extracted 83 F0 features (e.g., mean, range) and 130 features capturing multiple prosodic aspects (e.g., energy, spectral) using openSMILE. To measure attention, infants’ direction of gaze was coded. The prosodic features were selected with regularized regression and refined using backward regression. We assessed model performance with cross-validations. Results showed that greater F0 variability and sharper F0 increases towards the end of the utterance predicted greater increases in the proportion of time infants gazed at the mother’s face or interaction-related objects following IDS. Among other features, Mel-frequency cepstral coefficients and harmonic-to-noise ratio were selected, suggesting that maternal vocal timbre and breathiness may also contribute to the regulation of infants’ attention in real time.