A real-time speaker-independent continuous speech recognition system based on demi-syllable units.