Automatic acoustic-phonetic analyzes of thousands of hours of conversational speech in hard of hearing children.
Mark VanDam · The Journal of the Acoustical Society of America · 2011
The current longitudinal project examines speech and language development in 40 preschool children with mild- to severe-hearing loss (and a cohort of typically developing peers) as part of large, multicenter project on outcomes of children with hearing loss. Whole-day recordings are collected once monthly for 1 year and subjected to detailed acoustic-phonetic analyzes. To date, we have collected several hundred whole-day recordings using a commercially available, body-worn recorder and software (the LENA Foundation). The software performs unsupervised, offline analysis of the audio and produces (1) a time-aligned, XML-coded output including the talker labels adult-female, adult-male, target-child, and other-child and (2) the PCM wave file of the entire day. Audio segments of specific talkers are then extracted with custom software designed for detailed acoustic-phonetic analyzes of speech. For example, fundamental frequency was extracted for target-child utterances and compared in varied conversational exchanges with different interlocutors. This technology is suited to examine large-scale natural speech and langauge corpora. Developmental changes can be examined in domains ranging from prosody to conversational use of speech. Details of this research project, including advantages and challenges, will be discussed as well as theoretical implications for phonology and phonetics of child speech development. [Work supported by: NIH/NIDCD 5R01-DC009560-01S1.]