Dynamic Range Variability of Sound Energy Decay for Reverberation Time Estimation in Railway Noisy Environments

In this paper, we are interested in real time estimation of reverberation time (T60) for public address systems used to broadcast vocal announces in noisy railway environment. As the T60 of a space varies when its occupancy level changes, dynamic estimation of this parameter is needed for several purposes including speech enhancement algorithms. In this work, the standard "sound energy decay" method for T60 estimation is considered. We specifically highlight the limitations of this method when used for in-vivo T60 estimation: the reduction of the sound decay length, the non-stationarity of the excitation signal, the ambient noise and the adaptive energy decay estimation combine to reduce the dynamic range used for the estimation of T60. An alternative is then proposed in order to predict the boundaries of the dynamic decay regarding in-vivo estimation conditions. A set of real impulse responses and recordings of railway station ambient noises are used to analyze and quantify this dynamic range variability.