That sounds like a really interesting problem. You'd need microphones good enough, small enough, and cheap enough to hear the signal, Signals processing/machine learning software robust enough to pick up a user-selected pattern and in a reasonable latency, and then a processor small enough and power efficient enough to process it all while again also being economical enough, and then have it last 8 hours.