How to discriminate speech from nonspeech ?

[xinhui zhou <zxinhui@xxxxxxxxxxxx>]

Dear all,

 I am trying to use independent component analysis (ICA)  with data
from two microphones to separate two sound sources. One source is
speech and the other source is background noise such as bombing sound.
However, ICA can not decide  which output is speech dominant and which
output is noise dominant.

 Is there any algorithm which can distinguish speech from nonspeech
or  determine  which  output from ICA is  speech-dominant ?

 I found some papers such as the one shown as below,  but it seems
very sophisticated.

Discrimination of speech from nonspeech based on multiscale
spectro-temporal Modulations
Mesgarani, N.; Slaney, M.; Shamma, S.A.
Audio, Speech, and Language Processing, IEEE Transactions on
Volume 14, Issue 3, May 2006 Page(s): 920 - 930

 Thanks,

Xinhui,