The aim of the paper is to develop a robust virtual sensing method where dynamic variations in the actual acoustic system would not degrade the sensing performance. The approach here is to consider the possible uncertainties in the systems and to take into account this information to develop a virtual sensing method that is robust against these uncertainties. A certain sensing performance can be enforced and the task is formulated as an optimal robust control problem that includes uncertainty modelling. Numerical studies are performed on an acoustic duct system with varying properties, which show a satisfactory performance of virtual sensing when the system varies within a particular range. The proposed approach guarantees a certain level of performance robustness for virtual sensing when the systems are expected to vary during operations. Therefore, the approach can be used for practical implementation in actual acoustic systems where it is possible that the systems might vary during the sensing and control operation.