We propose a novel method to real-time in-cylinder pressure estimation by processing structure-borne sound measurements. It has been shown that knowledge of the in-cylinder pressure opens the door to robust misfire detection and sophisticated closed loop engine control schemes. However, the costs of such sensors have inhibited their use in production engines. On the other hand, acceleration sensors are of low cost and already mounted on modern production engines for knock detection. Since structure-borne sound is measured on the surface of the engine, all cylinders are simultaneously observed by one sensor. A simple physically based model, describing the speed dependence of the transfer behavior from each in-cylinder pressure to structure-borne sound is developed. Based on this model, a method for identifying the parameterized transfer function speed independently is developed. Once this algorithm was applied, pressure signal approximations can be extracted from structure-borne sound signals. Since the identification was performed speed independently, only one set of system parameters is needed. Experiments based on real measurement data show the capability of the proposed methods.