This paper is focused on the triggering of Spontaneous Head Movements (SHM), i.e. head movements which are supposed to be used to get additional information on a specific area of the robot environment. For that purpose, a Dynamic Weighting model (DWmod) is formulated as a low-level attention algorithm which allows an exploratory robot to drive its attention toward important items. DWmod is primarily based on auditory information, possibly coupled with visual data. These audiovisual characterizations rely on classification experts whose outputs are used by DWmod to trigger a SHM. The attention mechanism modeled by DWmod is rooted in the notion of congruence and predictability of items, allowing the robot to dynamically create its own rules about what is important or not in the current scene. This behavior is especially relevant in exploration tasks and particularly in search & rescue scenarios, where the robot has to react quickly with potentially no knowledge at all about the current environment. Within the Two!Ears framework1, a three-layered architecture on a simulated robot is developed, and simulation results demonstrate how the proposed model outperform basic low-level auditory-based turn-to reflexes.