Conference Proceedings

Abstract

The seemingly mature technologies for public address systems are not often as straightforward as they are thought to be. The usability of such systems depends, to some extent, on their capability to adapt to ambient noise so that the required intelligibility can be achieved, while unnecessary loudness is avoided for the tranquility of environmental sound. The time varying and unpredictable nature of noise in occupied spaces necessities the deployment of intelligent adaptation. A new scheme to achieve a specified speech transmission index, an objective acoustic parameter for speech intelligibility, is developed in this paper based on a set of blind estimation algorithms using machine learning. The paper details the rationale of the method and associated algorithms. In addition, for systems designed to achieve a specific signal to noise ratio, a simplified version is derived.