Phoneme-based continuous speech recognition without pre-segmentation