Efficient modeling of room acoustic impulse response by segmentation into early and late components

Abstract

A Room Impulse Response (RIR) has a complex time-frequency structure that can be divided into a number of discrete early reflections followed by a reverberant tail. Many acoustic signal enhancement applications, such as dereverberation and room equalization, require simple yet accurate models to represent an RIR. Parametric modeling of RIR typically approximate the overall response for a given position of a source and receiver inside a room, without considering the fundamental distinction between early reflection (modal region) and late reverberation (statistical region) of an RIR. In this paper, the RIR is estimated in two steps: first, the early reflection component is modeled utilizing fixed pole, parallel orthonormal basis functions to represent the modal behavior of the room. The late reverberation component of an RIR is more statistical in nature with exponentially decaying envelope, exhibiting different decay time constants for different frequencies. Wavelet decomposition using a multi-rate analysis filter bank is applied to model this part of the RIR. An iterative method is used to estimate the parameters of the model from a measured target RIR. The proposed hybrid method provides an accurate representation of an RIR while requiring a smaller number of parameters in comparison to existing methods.