The first ideas of using machines to replicate speech were in the early nineteenth century, but it wasn't until Bell Labs produced the Voder in the 1930s that they became a reality. This needed an operator, however, which somewhat invalidated the use of the machine, since with an operator, you could just employ them to speak themselves! However, by the 1960s, the systems had become automatic. They were employed to help the blind read, and for use in telephony, and by 1978 were introduced to children in the form of Texas Intruments' Speak and Spell.
Modern systems are attempting to improve the accuracy of the synthesis further to make the sounds more realistic, and are getting towards the stage that it is very hard to tell the difference between a human and a simulation.