you are correct.
add a dedicated program to process any gathered audio into text... and open the mic for constant sampling. once the audio is in text format, route it to processing to be indexed and scored against a phrase matching database. anything above a certain level would be automatically routed to the attending agent.
and yes, i could implement such a system
Looked at your FR homepage... I like it :-)
What do you do? I’m in embedded systems.
Remember that a smartphone can be operated out-of-band.
The normal cell frequencies are far too cluttered to be useful for surreptitious monitoring... sending data out as normal cell data is far too risky for serious work.