A new approach to the permutation problem for Blind Source Separation (BSS) in the frequency domain is presented. By means of a state-space representation, the alignment is reduced to a recursive adaptive tracking of state trajectories associated with the demixing matrices. The estimated smooth trajectories are used to initialize the Independent Component Analysis (ICA) in order to force it to converge with a coherent permutation across the whole spectrum. Since permutations are solved with no information about the signal power, this method works also for short utterances (0.51 s) and in highly reverberant environment (T60 ~ 700 ms). Furthermore it is shown that the underlying frequency link, provided by the recursive state estimation, increases the accuracy in the ICA step when only few observations are available.
A BSS method for short utterances by a recursive solution to the permutation problem
Nesta, Francesco;Omologo, Maurizio;Svaizer, Piergiorgio
2008-01-01
Abstract
A new approach to the permutation problem for Blind Source Separation (BSS) in the frequency domain is presented. By means of a state-space representation, the alignment is reduced to a recursive adaptive tracking of state trajectories associated with the demixing matrices. The estimated smooth trajectories are used to initialize the Independent Component Analysis (ICA) in order to force it to converge with a coherent permutation across the whole spectrum. Since permutations are solved with no information about the signal power, this method works also for short utterances (0.51 s) and in highly reverberant environment (T60 ~ 700 ms). Furthermore it is shown that the underlying frequency link, provided by the recursive state estimation, increases the accuracy in the ICA step when only few observations are available.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
