Michael A. Casey
We present results from two experiments that show how a multi-layer neural network can be used to estimate parameters for non-trivial musical instrument physical-modeling systems. In the first experiment a connectionist model was trained to estimate four parameters to a brass model and in the second the network was trained to classify and estimate four parameters to two physical models; a trumpet and a saxaphone. We present these results in the context of continuing research into human timbre perception and classification.