. "For applications such as speech recognition, however, if the threshold is set high enough that all background noise softer than a word is discarded, then the samples that immediately proceed the first over-threshold sample may in fact contain information that would aid the recognition of that word." . . . .