The human hearing system is quite insensitive compared to many wild animals. A sensitivity better than the Brownian noise is of little value. However, the human threshold is at least 20 dB worse than the Brownian noise. For weak audio recording, aiming for just the human threshold of hearing is not enough. The micrphfone+amplifier noise should instead be slightly lower than the equivalent Brownian noise.
There are two reasons for matching source and load in RF.
1.) reflections: When cable is longer than 1/10 and definitely longer than 1/4 wavelength, the standing wave issues should be considered. At 20 kHz and considering velocity factor, this is in the ballpark of 1000 to 2500 m. Not relevant for baseband audio, but a serious issue for ADSL tones (fmax 1 MHz or 2 MHz in ADSL2+). 2.) You do not want to throw away any available power for best noise match. While in practical RF preamplifiers the lowest noise is obtained when the source and load are relatively closely matched. The best noise match is in the ballpark of power match but not always exactly.In a power match situation, the source terminal voltage is half (-6 dB) compared to source open circuit voltage. Unfortunately, you can't gather any information from the source in open circuit condition unless the load impedance is very high.
Assume a low impedance dynamic microphone connected directly to the high impedance grid of a pentode. The pentode will have all kinds of voltage noises of its own. The weak microphone signal voltage will not override the pentode noise voltage, but a lot of microphone current would be available, but is not used.
If we put a step up transformer in between and the secondary voltage will now better override the pentode noise and more source signal current is also used. This improves the SNR.
At power match, the transformer primary is -6 dB below open circuit voltage. At 10 times load impedance, the primary voltage would be only
-0.8 dB down.
In a power match with say 100 ohm source impedance, a 1:10 impedance transfer ratio and 1000 ohm load, the voltage across the secondary would be 3.1 times (+10 dB) the primary voltage. Consider the source loading loss, the net gain would be +4 dB compared to source unloaded voltage.
In the lightly loaded condition Rl = 10Rs with a 1:1 transformer and
1000 ohm load resistor the resistor voltage would be that -0.8 dB down compared to source open output voltage.Comparing the two cases, the power matched condition will give about 5 dB stronger voltage across the 1000 ohm load resistor. If the amplifier own noise voltage remains constant, the power match has 5 dB better SNR. .
If power match gives 5 dB better SNR, some equalization losses may still be tolerated.