A heart rate of a subject is determined using a probabilistic segmental model. Reflections of ultrasound signals from the subject, for instance, are down modulated to an audio band and velocity samples are obtained. For a signal analysis window of the samples, e.g., 25 msec. of samples, features of the samples are obtained. Based on the features, a first set of probabilities is obtained for different candidate current heartbeat periods. A second set of probabilities is determined based on combinations of different previous heartbeat periods. One of the candidate current heartbeat periods is determined to be most probable, based on the first and second sets of probabilities.