breathing sounds when using a grammar model for speech recognition

[alexkinren]

"I am using Julius for speech recognition with the following grammar configuration:
julius -C /usr/local/src/julius/julius-kit/dictation-kit/am-gmm-monophone.jconf -walign -nostrip -1pass -gram /usr/local/src/julius/dict_test/test -input file.
I've encountered an issue.
When there is a breathing sound lasting about 500ms in the audio file, all the subsequent speech is recognized as [/s]. Do you have any solutions for this?