Extracting room reverberation time from speech using artificial neural networks

A novel method to extract the reverberation time from reverberated speech u tterances is presented. In this study, speech utterances are restricted to pronounced digits; uncontrolled discourse is not considered. The reverberat ion times considered are wide band, within the frequency range of speech ut terances. A multilayer feed forward neural network is trained on speech exa mples with known reverberation times generated by a room simulator. The spe ech signals are preprocessed by calculating short-term rms values. A second decision-based neural network is added to improve the reliability of the p redictions. In the retrieve phase, the trained neural networks extract room reverberation times from speech signals picked up in the rooms to an accur acy of 0.1 s. This provides an alternative to traditional measurement metho ds and facilitates the occupied measurement of room reverberation times.