When building a system of automatic speech recognition, there is problem of reverberated speech correcting, for which solution it is necessary first measure a reverberation time. Blind measurements of reverberation time are less efficient, in the sense of speech recognition accuracy, compared with direct measurements, but it is hard to find in the literature estimations of this loss. In this paper, this gap is replenished and conditions, the implementation of which helps to minimize this loss by reducing it to 3-5%, are defined.
reverberation time measurement; blind measurement; automatic speech recognition; speech recognition accuracy
"Slepoe yzmerenye vremeny reverberatsyy v systemakh avtomatycheskoho raspoznavanyia rechy" ,
Information Processing Systems,