Optimizing the recognition lexicon for automatic speech recognition by