Vocal Biomarker Assessment Following Pediatric Traumatic Brain Injury: A Retrospective Cohort Study 1 MIT Lincoln Laboratory, Lexington, Massachusetts, USA 2 Stanford University, Stanford, California, USA 3 Mass General Hospital, Boston, Massachusetts, USA 4 University of Texas, Dallas, Texas, USA cnoufi@stanford.edu, [adam.lammert, daryush.mehta, jrw, gregory.ciccarelli, sturim]@ll.mit.edu, jgreen2@mghihp.edu, thomas.f.campbell@utdallas.edu, quatieri@ll.mit.edu Abstract Recommendations following pediatric traumatic brain injury (TBI) support the integration of instrumental measurement to aid perceptual assessment in recovery and treatment plans. A comprehensive set of sensitive, robust and non-invasive mea- surements is therefore essential in assessing variations in speech characteristics over time following pediatric TBI. In this paper, we discuss a method for measuring changes in the speech pat- terns of a pediatric cohort of ten subjects diagnosed with severe TBI. We apply a diverse set of both well-known and novel fea- ture measurements to child speech recorded throughout the year following diagnosis. We analyze these features individually and by speech subsystem for each subject as well as for the entire cohort. In children older than 72 months, we find highly sig- nificant (p< 0.01) increases in pitch variation and number of unique phonemes spoken, shortened pause length, and steady- ing articulation rate variability. Younger children exhibit sim- ilar steadied rate variability alongside an increase in articula- tion complexity. Nearly all speech features significantly change (p< 0.05) for the cohort as a whole, confirming that acoustic measures expanding upon perceptual assessment are needed to identify efficacious treatment targets for speech therapy follow- ing TBI. 1 Index Terms: pediatric, traumatic brain injury, vocal biomark- ers, dysarthria, dysphonia, speech subsystems, acoustics, longi- tudinal, retrospective, cohort study 1. Introduction The recovery process following severe traumatic brain injury (TBI) in children varies widely. Each individual may experience a combination of cognitive, perceptive, emotional and motor function impairment [1]. Voice and speech disorders commonly occurring post-TBI indicate impairment to the brain’s widely distributed speech network. Commonly reported characteristics of speech altered by dysphonia, a laryngeal function disorder, 1 Approved for public release. Distribution is unlimited. This mate- rial is based upon work supported by the Under Secretary of Defense for Research and Engineering under Air Force Contract No. FA8702- 15-D-0001. Any opinions, findings, conclusions or recommendations expressed in this material are those of the author(s) and do not neces- sarily reflect the views of the Under Secretary of Defense for Research and Engineering. This research was supported, in part, by the Na- tional Institute on Deafness and other Communication Disorders Grant R01DC0368. and dysarthria, a motor-speech disorder, are found within the articulatory, prosodic and phonatory speech subsystems [1–4]. A closer look at these affected speech subsystems can provide refined cues to impairments of the brain’s speech network as well as identify efficacious treatment targets for speech therapy [4]. Recommendations following pediatric TBI support the in- tegrated use of objective, instrumental measurement alongside perceptual assessment for informing and aiding in recovery and treatment plans [5]. Vocal biomarkers used to monitor and pre- dict cognitively-impaired speech have been studied for several decades [4] and many commercial acoustic analysis programs are available [2]. However, to our knowledge, acoustic analy- sis has only been applied to the longitudinal study of speaking rate following pediatric TBI [6]. In this article, we define and apply a diverse set of acoustic, audio-based vocal biomarkers to study TBI-affected speech of a pediatric population over time. We apply a diverse set of both well-known and novel speech feature measurements to speech recordings taken monthly over the course of one year and summarize the longitudinal changes in these features via a trend profile. This allows us to track how particular characteristics of each speech subsystem are chang- ing in relation to each other. Section 2 details this methodology. In Section 3, we analyze the trends exhibited by the cohort and comment on the nuanced relationship between recovery trends and age. In Section 4, we discuss the applicability of this fea- ture set in assisting assessment throughout recovery. Section 5 concludes our work. 2. Methods 2.1. Study Design and Data Collection We apply our feature analysis to speech data originally collected by Campbell et al. to understand longitudinal changes in the perceptual ‘Percent Consonants Correct - Revised’ (PCC-R) in- telligibility measure. This data consists of recorded conversa- tions between a subject and a trained examiner [7]. Due to the large amount of preprocessing required to utilize this dataset for acoustic analysis, we focus on ten out of fifty-six subjects within the original study. The ten-subject cohort was selected to be representative of the age range within the original sam- ple, include both males and females, and have corresponding recordings of good perceptual quality. Each subject had been recently diagnosed with a severe Camille Noufi 1,2 , Adam C. Lammert 1 , Daryush D. Mehta 1 , James R. Williamson 1 , Gregory Ciccarelli 1 , Douglas Sturim 1 , Jordan R. Green 3 , Thomas F. Campbell 4 , Thomas F. Quatieri 1 Copyright  2019 ISCA INTERSPEECH 2019 September 15–19, 2019, Graz, Austria http://dx.doi.org/10.21437/Interspeech.2019-1200 3895