I suggest you read the AppNotes from Analog Devices Incorporated. As a kid, I learned much useful analog/signal_processing ideas and maths and building ideas from their freely available publications.
The Analog Dialogue was one which I received monthly. So google that, and start reading, and perhaps BUILDING real circuits.
Use an old loudspeaker, from anold transistor radio, as your microphone.
You might reverse_wire the tiny transformer, to stepup the voltage to be a useful level.
You can touch the wires to your earbuds, and hear your fingernail scratching across the loudspeaker's paper cone.
Get starting learning.
To be concrete about discerning important output signals from sensors ----- theories exist on Matched_Filtering, to optimally extract signals from noise or to optimally discriminate between TWO choices of possible signals.
Just consider the integral, over a complete period, of [ sin(x) * cosine(x)]
The integral is ZERO, when integrated over N * period_duration.
Such signals (sin, cosing) are deemed orthogonal.
Many other "signals" have been defined, to achieve orthogonality.
Some of these "signals" look like noise (randomness) yet to the crafted receiver of such signals, able to implement correlators, a strong and robust datalink exists.
Peter Mallory did much work in this area.