From the book I’m reading, An Introduction to Psycholinguistics:

DeCasper and Fifer (1980) reocrded mothers reading a story to their newborns.  Then, their 3-day-or-younger infants were given a pacifier connected to a computer that would play recordings of the mother’s voice or of another woman’s voice.  A high rate of sucking on the pacifier would activate the playing of the mother’s voice.  Comparing changes on the sucking rate with the infant’s baseline rate, the researchers found that the infants sucked more in order to activate the tape with their mother’s voice than to hear the voice of another woman!

The requirement was then changed so that the infants had to suck at a lower rate than normal in order to hear their mother’s voice.  The infants quickly changed to slower rates, thus demonstrating that they could distinguish sound of their mother’s voice and that of another woman.  Locke (1993), however, suggests the learning of the mother’s voice may actually have occurred, not prenatally, but within the first 12 hours after birth when the mother was talking to the newborn.  Since the measurements were taken after the 12-hour period, this could well be the case.  If so, then there may not have been prenatal learning.

I think equally as astounding is that a baby of 3 days or less is able to learn and respond based on controlled stimulation.  Just amazing!