While text-to-speech has been around since before the Internet, it has taken some time to actually reproduce a real speaker's voice. Now, three seconds of speech is all it takes.