Project 2 - Pediatric Production Training Software
H. Timothy Bunnell, Ph.D. - Wilmington, DE / A. I. duPont Hospital for Children
Purpose of Study
The purpose of the study is to extend Software called STAR (Speech Training and Auditory Rehabilitation) to assist with hearing and speech habilitation for children who have received Cochlear Implants. The software uses utterance verification technology within the context of a computer game to elicit speech and provide feedback on the extent to which utterances match age-typical norms for speech acoustics. The system is intended to work as an adjunct to conventional A/V therapy by automatically providing drills, objectively monitoring progress, and assisting in record keeping and reporting.
The first two years of the project will be spent collecting baseline data necessary to design and refine the infant utterance verification technology used in STAR. The data will be obtained by recording audio and video therapy sessions for participating CI children. From these recordings, we will extract and label all usable speech-like vocalizations (canonical syllables, words) made by the children along with a representative sample of non-speech vocalizations (coos, squeals, laughs, cries, etc.). From these data we hope to develop acoustic models of CI infant vocalizations during this crucial developmental period. These models will be used to inform the utterance verification technology for the STAR system. We will also examine the development of vocalizations over time to identify the order and rate at which verbal linguistic skills emerge for each child and the relationships among factors such as type and complexity of vocalization, type of implant, mapping, chronological age, age of implantation, etc.
Simultaneously with baseline data collection, we plan to develop the existing STAR software to make the system suitable for perception as well as production training. Years three through five of the project will then be devoted to formative testing and revision of the system, first in the clinic, and later as a take-home system for use between therapy sessions in a home setting. At the conclusion of the project, we anticipate a fully developed and tested system, ready for distribution and/or commercialization.
Audio and video recordings were collected from nine children with CIs during their weekly speech therapy sessions. Speech and non-speech vocalizations from the recordings were labeled by hand and used to inform models for automatic labeling. Speech-like vocalizations from one child with CIs collected over ten months (age 21 to 30 months) were labeled with automatic speech recognition based on Hidden Markov Models and acoustically analyzed. The preliminary results indicated the child expanded her speech repertoire (Nagao, Cornaglia, and Bunnell, 2010a, 2010b).
Recent Publications and/or Presentations
H. Timothy Bunnell
Email - firstname.lastname@example.org
Phone - (302) 651-6835