For sounds composed of many densely packed modes, a modeling of each mode is impractical.
In such a case, one should choose only the most perceptually important modes to model as above. The remaining densely packed modes can then be statistically modeled using exponentially enveloped noise that is filtered by a lowpass filter whose bandwidth is decreasing over time.
The time-varying lowpass filter causes the higher frequency energy in the noise to decay faster, a natural property in most struck percussive objects.
A further refinement can include a strike-dependent lowpass filtering to simulate mallet brightness.
A block diagram of the full CMS algorithm with statistical mode modeling is shown in Fig. 17.
Figure 17:
The CMS algorithm with statistical mode modeling.