engleski

An Improved Algorithm for Maximum-Likelihood-based Blind Estimation of Reverberation Time

Reverberation is a well known phenomenon in the area of room acoustics and speech processing representing the gradual sound energy decay in an enclosure after the sound source has been switched off. Due to detrimental effects of temporal and frequency smearing of the speech signals received on microphones that it causes, numerous methods for signal enhancement have been proposed, where a number of them require the reverberation time (RT) parameter value to be known in advance. In this paper an improved algorithm for blind time-domain estimation of reverberation time from received speech signals is presented, where novel rules for defining the optimal boundaries of the speech segments used for maximum-likelihood estimation are introduced. We examine the performance of our algorithm, more precisely, the deviation of the estimated RT values and their variance from the true value of the parameter, and show that the estimation accuracy of our algorithm over similar methods is improved over a wide range of reverberation time values.

reverberation time; blind estimation; speech; maximum-likelihood; coefficient of determination

nije evidentirano