Although language delay is common in autism spectrum disorder (ASD), research is equivocal on whether speech development is affected. We used acoustic methods to investigate the existence of sub‐perceptual differences in the speech of toddlers who developed ASD. Development of the distinction between b and p was prospectively tracked in 22 toddlers at low risk for ASD (LRC), 22 at high risk for ASD without ASD (HRA−), and 11 at high risk for ASD who were diagnosed with ASD at 36 months (HRA+). Voice onset time (VOT), the main acoustic difference between b and p, was measured from spontaneously produced words at 18, 24, and 36 months. Number of words, number of tokens (instances) of syllable‐initial b and p produced, error rates, language scores, and motor ability were also assessed. All groups' mean language scores were within the average range or slightly higher. No between‐group differences were found in number of words, b's, p's, or errors produced; or in mean or standard deviation of VOT. Binary logistic regression showed that only diagnostic status, not language score, motor ability, number of words, number of b's and p's, or number of errors significantly predicted whether a toddler produced acoustically distinct b and p populations at 36 months. HRA+ toddlers were significantly less likely to produce acoustically distinct b's and p's at 36 months, which may indicate that the HRA+ group may be using different strategies to produce this distinction. Autism Res 2017. © 2017 International Society for Autism Research, Wiley Periodicals, Inc. Autism Res 2017, 10: 1269–1279. © 2017 International Society for Autism Research, Wiley Periodicals, Inc.