Physical modelling has been broadly used in the past decades to analyse the function of musical instruments. It offers the possibility to synthesise sounds by simulating the oscillations of the instrument and link certain instrument characteristics to the produced sound. Recent studies aim at characterising the actions of the player using experimental measurements combined with physical models. The current study uses an instrumented clarinet mouthpiece to obtain data during musical performance. Based on the measured signals, an inverse modelling approach is established that attempts to resynthesise the recorded sound. Particular emphasis is given on tonguing techniques at note transitions, in order to simulate the articulatory actions of the player. To this end three nonlinear components are used to model the excitation mechanism, offering extensive control over the virtual resonator. The inverse model consists of an optimisation routine that searches through a parameter space in order to estimate the values of the physical parameters that can resynthesise the target signal. To ensure algorithm convergence, special attention is given on both efficiency and stability while formulating the physical model. Regarding numerical stability, an energy balance of the simulated system is considered. Regarding computational efficiency, a one-dimensional time-domain resonator model is adopted, that is coupled to a lumped excitation mechanism which models the reed-mouthpiece system, along with any actions of the player's embouchure. The inverse model is capable of resynthesising an excerpt of Weber's clarinet concerto No. 2, with the simulated sound being qualitatively similar to the recorded one.
