Leaks information about the voice stream. It's not inconceivable that a well trained algorithm could recover sentences from the transmission pattern, although I don't know for sure if it could do quite that well.