Speech Dereverberation Using Linear Prediction

摘要

This article describes a method of reducing reverberation in an observed signal to mitigate problems experienced in speech communication systems, such as hands-free mobile telephones, videoconferencing systems, hearing aids, and voice-controlled robots. It focuses on dealing with the effect of late reverberations, which are known to be a major cause of the degradation of automatic speech recognition (ASR) performance and loss of speech intelligibility. Experimental results show that this method can provide substantial improvements in ASR performance and audible quality under severely reverberant conditions.