I first encountered this about 20 years ago when chatting to telecoms R&D peeps. They were doing some very fine measurements of a digital protocol on a fibre loop on their campus. The noticed a signal superimposed on the digital bitstream. It wasn't high enough to result in BER, but it was there nonetheless. After checking it wasn't an error in the transmitter they concluded it was being injected, so they demodulated it and heard a pneumatic jackhammer! They found some workmen near one of the manholes.
They decided that sound causes a longitudinal change in the fibre, which affects the light carrying properties.