Estimation of Reverberation Time by Performing Acoustic Echo Cancellation Considering Near-end and Far-end Speech Signals

Abstract

In teleconferencing, there will be both Near-End (NE) speech and Far-End (FE) speech signals. The FE speech signal bounces inside the room and mixes with the NE speech signal and creates echo during the conferencing. This echo is known as the acoustic echo. There are many methods to mitigate the acoustic echo. This paper adopts LMS algorithm for the acoustic echo cancellation (AEC). The echo return loss enhancement is also performed to find out the quality of echo present in the speech signal after the AEC. The reverberation time of the residual echo embedded in the speech signal is calculated using two methods and those values are compared. The methods adapted are maximum likelihood detection, also known as the online estimation method and the Schroeder method. Various cases of the room impulse response characteristics are considered from Aachen Impulse Response (AIR) database. This paper brings out the idea of estimation of reverberation time by performing AEC. It proves that the error percentage obtained after performing AEC is less than that obtained before performing AEC. The proposed logic gives the reverberation time value of 1.13 s with AEC, which was previously 1.20 s without AEC. This clearly shows that there is 0.07 s reduction in the reverberation time.