Fig. 5
From: A neural network-supported two-stage algorithm for lightweight dereverberation on hearing devices
Comparison of the true RIR (full line) vs. the estimated RIR (dashed line). Estimated linearized transfer function of the system which applies O-PSD-WPE on reverberant speech is shown as a dotted line. We observe strong dereverberation in the given filter range [\(\Delta , \Delta + \tilde{K}-1\)] and shortly afterwards because of recursive averaging. Only the \(\textrm{T}_{30}\) range is displayed as it is the valid estimation range for the estimated RIR. \(\textrm{T}_{\textrm{60}} = 0.8s\). HA scenario [ \(\varvec{\nu } + \varvec{d} + \varvec{e}\) ; \(\Delta = \tilde{\Delta } = 5\) ]