It's also the case that often there aren't spaces between words. One of the surprising things when you first start doing audio analysis is that the pauses between words that you perceive aren't actually there in reality -- which makes them a lot more difficult to 'remove' algorithmically.