We are tuned to hear words amongst noise, it's how we can follow two different conversations in a noisy pub while there is music playing, and other conversation all around us. We can select and hone in on specific voices. This takes a lot of brain power. When that ability is focussed on white noise, it will occasionally have a false positive, because it's searching for a meaningful signal, and if you look hard enough in white noise, you'll always find the signal you are after (this is physics, please go read).
On a related note, does anyone remember the Bullshit law suits in the 80's against various Heavy Metal bands, for allegedly hiding messages in reverse on their albums?
This is the same phenomena. A sound source interpreted incorrectly as certain words, when we know full well what is actually is (reversed different words).
Now, apply that to something as complex as white noise. There's plenty of stuff going on there, you just have to filter out certain bits, focus on others, and listen to enough, and part of it will sound like speech. It's the 'million monkeys' analogy at work. Except you only get phrases, not whole Shakespeare plays.