[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: CASA problems and solutions


While I wouldn't want to prejudge the usefulness of your approach as a
method for scene analysis by machines (I've heard some impressive
reverberation-suppression demos of Birger Kollmeier's), it seems
evident that humans do not use the cues in the way you describe.
Firstly, although human sound localisation works quite well in
reverberant environments, humans have great difficulty using
differences in source location to separate speech in reverberation
(Plomp, 1976; Culling et al., 1994; Darwin and Hukin, 2000). Second,
there is some reason to doubt that sound localisation is a pre-cursor
to sound separation (Culling and Summerfield, 1995; Hukin and Darwin,
1995). Darwin and Hukin (1999) have suggested that separation occurs
first, and that the processes of sound localisation are then applied
to already-separated auditory objects.


Plomp, R (1976). Acustica, 34, 200-211.
Culling, JF, Summerfield, Q, and Marshall, DH (1994). Speech Comm. 14, 71-95.
Darwin, CJ and Hukin RW (2000) JASA, 108, 335-342.
Culling, JF and Summerfield, Q (1995) JASA 98, 785-797.
Hukin, RW, and Darwin, CJ (1995) JASA 98, 1380-1387.
Darwin, CJ and Hukin, RW (1999) JEP:HPP 25, 617-629.