A Linear Model of Speech Timing