On the Efficiency of Speech Sounds