In this work we present an extension of the classic schema of a time-varying filter excited with white noise for the modeling of noise signals from musical instrument sounds. The framework used is that of statistical signal processing, and a structure that combines an Autoregressive (AR) model with an adaptive FIR filter is proposed. A combbased structure for the AR filter is used when tuned noise is to be modeled. The analysis/resynthesis schema proposed is used to perform some basic sound transformations such as time stretching, tuning and energy envelop control, and spectral processing.