The ability to predict when something will happen facilitates sensory processing and the ensuing computations. Building on the observation that neural activity entrains to periodic stimulation, leading neurophysiological models imply that temporal predictions rely on oscillatory entrainment. Although they provide a sufficient solution to predict periodic regularities, these models are challenged by a series of findings that question their suitability to account for temporal predictions based on aperiodic regularities. Aiming for a more comprehensive model of how the brain anticipates 'when' in auditory contexts, we emphasize the capacity of motor and higher-order top-down systems to prepare sensory processing in a proactive and temporally flexible manner. Focusing on speech processing, we illustrate how this framework leads to new hypotheses.