Detection and estimation of arrivals in room impulse responses by greedy sparse approximation

We investigate the use of greedy sparse approximation for facilitating the time-domain analysis of room impulse responses (RIRs), specifically locating the times and amplitudes of arrivals to not long after the upper bound of the “mixing time,” i.e., the time after which there exists in theory the same number of sound rays per unit volume throughout the room. We compare the performance of two methods of greedy sparse approximation - matching pursuit (MP) and orthogonal MP (OMP) - for estimating arrival times and amplitudes. By using RIRs generated from a stochastic model, we quantify the performance of each estimator using dynamic time warping to optimally pair estimated and true arrivals. We find OMP significantly outperforms MP in estimating both the arrival times and amplitudes, and having fewer erroneous and duplicated arrivals.