Noise Power Spectral Density Estimation for Public Address Systems in Noisy Reverberant Environments

In many public areas, public address systems play a central role in broadcasting important messages. Consider e.g., airports or train stations. The intelligibility of messages broadcast by such public address systems gets reduced due to environmental noise sources. Usually, these systems therefore employ a pre-processing module to modify the original message such that its intelligibility in the noisy environment is increased. Most of the pre-processing methods for intelligibility improvement require an estimate of the environmental noise PSD. However, existing noise PSD estimators overestimate the noise PSD in reverberant environments, leading to an overamplification of the speech by the speech intelligibility pre-processing module. In this paper, we propose a new MMSE-based noise PSD estimator that can be utilized in public address systems considering reverberation.