Hidden semi-Markov models are traditionally exploited in discrete time. Here we illustrate an application where hourly heart rates are segmented by a hidden semi-Markov model in continuous time. The model is a mixture of Gaussian distributions, whose parameters evolve according to a latent semi-Markov chain in continuous time. The chain switches between two states and the time the chain dwells in each state is tuned by covariates via an accelerated failure time regression model.