Parameter estimation and inversion for wind instrument physical models.
Mark J. Sterling, Xiaoxiao Dong, Mark F. Bocko · The Journal of the Acoustical Society of America · 2008
An outstanding goal in the field of musical instrument physical modeling has been the definition of a robust parametric and expressive representation or coding for musical audio. For this purpose, accurate dynamical models of musical instrument systems must be supported by algorithms that can provide suitable sets of control parameters for producing some desired output. We describe an estimation algorithm for clarinet physical modeling. However, the results could be applied to a broader class of convolutional wind instrument models, which fit the McIntyre–Schumacher–Woodhouse framework. Three fundamental signals in clarinet physical modeling are the blowing pressure pm, bore pressure pb, and flow u inside the mouthpiece. These are related via the nonlinear reed-valve characteristic and the air column impulse response. Assuming pb is given or approximated from a digital recording using known radiation characteristics, pm and u are iteratively estimated against one another, fitting the model to pb. This is accomplished with an informed inversion of the reed valve and an adaptive (LMS) inverse of the bore impulse response. This method supplants prior work we have done where blowing pressure waveforms were computed by envelope detection. Results are verified with numerous audio examples. [Work partially supported by NSF Grant IIS-555457.]