Analog Reservoir Computing using Microphone?

Reservoir computing is a neural-network based machine learning system where a nonlinear stage creates outputs that are trained on by classifiers.

Analog reservoir computing has been done using buckets of water and optoelectronics.

My question is, is there a way of doing it using a microphone. You feed a signal out through a speaker, the nonlinearities are introduced by the environment and then this is picked up by the microphone on the computer. Has this been tried?

Also, is there a material that can retain it's "memory" longer for this - something that is affected by the sound and thus retains a memory and also produces sound. A gel material for example. If anyone has ideas, that would be great thanks.