Noise-Robust Reverberation Time Estimation Based on Sound HOS Decay

This work is a part of a European project entitled I'CityForAll which aims to improve the mobility of elderly people suffering from presbycusis. In particular, this project proposes algorithms dedicated to public address systems, in order to enhance the intelligibility of vocal announcements broadcasted in noisy and reverberant public confined spaces such as railway stations. As the reverberation time (T60) of such spaces varies when the occupancy level changes, a dynamic estimation of this parameter is needed to optimize the performances of these algorithms.In this paper, we deal with a robust method for the estimation of T60 in noisy conditions. It is an approach which generalizes the standard sound energy decay method based on diffusing a noise sequence and performing the T60 estimation from its energy decay. However, in this novel method we rather propose to diffuse a signal having specific statistical properties and to estimate the T60 using its HOS (Higher Order Statistics) decay. The performances of this method are evaluated using a set of synthesized and real impulse responses, for realistic signal-to-noise ratios ranging from 5dB to 20dB.