Benchmarking early childhood spoken engagement in Hawaiian Preschool settings using Advanced Speech Processing and Sensing Technology

Capturing child–teacher communication in naturalistic classroom settings offers a valuable window into early engagement and learning dynamics. This study presents preliminary results from pilot data collected in a Hawaiʻi public preschool serving predominantly Pacific Islander, Filipino, and Native Hawaiian children. Speech data was collected during classroom activities using LENA, a wearable audio recording device worn by both adult and child participants. Pilot data was processed using the CRSS Speech and Speaker Processing Pipeline, developed with similar adult-child preschool interactions. Our current in-house speaker diarization pipeline achieves an accuracy of ∼78–82% across speaker classes (primary child, secondary child, adult), while automatic speech recognition (ASR) achieved a performance Word Error Rate of ∼20–40% for a sample range of child–adult conversational audio streams. Spatial proximity data (e.g., child-to-teacher distance) was collected using wearable Bluetooth Low Energy (BLE) beacons. BLE signals enabled spatial mapping of children’s proximity towards high-density linguistic zones—such as teacher-led stations—allowing for the construction of individualized engagement profiles. The collected naturalistic adult–child speech was also aligned with proximity data for child–teacher engagement tracking. This initial work demonstrates the potential of acoustic speech technology and proximity sensing to inform models of engagement in multilingual, culturally rich early learning contexts.