Abstract The goal of this R03 is to characterize longitudinal vocal development in infants and toddlers at risk for ASD and use these metrics to predict subsequent diagnosis and dimensional social/language abilities. There is a critical need for reliable markers of autism spectrum disorder (ASD) that can be used to detect the condition in infancy and hasten the onset of early intervention services. Preliminary data from our team indicates that infant vocalization features distinguish groups beginning in the first year of life, and account for significant variance in later diagnostic status and social/language phenotype. For example, we found that infants later diagnosed with ASD produce fewer speech-like vocalizations, fewer vocalizations directed toward others, more crying, and altered vocalization acoustics. However, the longitudinal development of these early-emerging differences has not been completely described, and while they hold promise as early markers, their ability to predict individual clinical outcomes has not been directly tested in a large sample. Therefore, the power of these early vocalization differences to improve clinical detection has not been fully harnessed. To address this gap, we leverage a large existing dataset collected through the Infant Brain Imaging Study (IBIS; NICHD R01HD055741, PI: Dr. Joseph Piven). In this successful longitudinal multi-site study, infants at high and low familial risk for ASD completed video-recorded behavioral assessments at 6, 12, and 24 months of age, as well as neuroimaging at the same time points. Our research plan includes annotating infant vocalizations produced during these interactions using an expanded version of a coding scheme we have already developed, manualized, and trained annotators to reliably implement, and measuring the acoustic properties of each annotated vocalization. In Aim 1, we will use mixed models to analyze developmental group differences in vocalization features, determining which qualities distinguish groups and when. In Aim 2, we will use machine learning to test whether vocalization features produced during the first year of life can accurately predict diagnostic and social/language outcomes at age 2. This work will provide evidence of the clinical utility of infant vocalizations as pre-diagnostic behavioral biomarkers, setting the stage for subsequent studies of the relationship between early vocal behavior and brain development (measured using previously collected neuroimaging data from the same participants). The