We set out to show that cognitive models of sound-generating systems could be used for identifying and classifying instrument sounds. We chose to focus on the estimation and classification of parameters from two different musical instrument groups; brass instruments and single-reed instruments. These were chosen because of the underlying similarity of their physical systems; if a system could be built to distinguish between these two groups of instruments then other groups would be considered less challenging.\
We propose that such estimation techniques could play an important role in the perception of timbre. The results of Grey [2] indicate that subjects tended to group instruments by family as well as by the psychophysical components of the signal. Our results suggest that such familial groupings could be obtained from the acquisition of models of sounding systems. Furthermore we have shown that it is possible to achieve reasonable articulator estimates for the sounding systems. This was another component of Grey's hypothesis; articulators were considered responsible for the exceptions to familial groupings in the data. In future work, it is our hope to demonstrate the importance of articulatory estimators for classification tasks.\
The multiple model example indicates how a number of learned models can be drawn from in order to classify a sound, and subsequently interpret the sound in terms of an underlying physical model and the articulator dynamics.\
We have shown that estimates can be obtained with no explicit inversions of the physical systems. Thus we believe the methods and techniques described in this paper to be cognitively plausible. However, future work will require the design and implementation of perceptual experiments in which the grouping preferences of human subjects are compared to the classification results of our inverse modeling systems.\