Robust speech/non-speech detection using LDA applied to MFCC for continuous speech recognition