> I see a lot of boards using caps as large as 470uf before BJTs. why so large? for better bass response?
Not at all. With typical 2K base resistors, 10uFd is -1dB at 20Hz, so 50uFd or 100uFd should be ample.
Noise power can be considered in terms of both noise-voltage and noise-current. Everything is a compromise: you select and bias the input device so the (noise voltage) summed with (noise current times source resistance) is a minimum. Tubes and FETs have very low noise current. Bipolar transistors that have been biased for low (lowest) noise voltage have quite high noise current.
Good clean transistors have flat ("white") noise spectrum from maybe 10Hz to up in the MHz. (At the high end, stability considerations in a practical amp may cause rising noise, though this should be small. At the low end, "1/f" noise is a problem, related to silicon cleanliness but probably fundamentally unavoidable.)
From the point of view of the input device, any input caps are part of the source impedance that noise-current makes noise in.
Say you have a 200Ω mike in series with a 10uFd cap. At 1,700Hz, 10uFd is like 10Ω reactance, plus 200Ω is 210Ω, "same as" 200Ω. At 170Hz, 10uFd is like 100Ω reactance, plus 200Ω is 300Ω, "a little more than" 200Ω. At 17Hz, 10uFd is like 1,000Ω reactance, plus 200Ω is 1,200Ω, "much more than" 200Ω. Current noise with a cap like that will rise in the bottom of the audio band.
Is that a problem? Large rise may dominate an objective measurement of noise, though you can legitimately apply an A-weight filter and ignore most of it. In many studios, ventilation noise is large in the lowest octave, so a little bass-noise in the amp may go unnoticed. And even without a cap and in a dead-quiet studio, transistor noise may be rising below 50Hz due to 1/f effects.
1/f noise is hard to measure and impossible to sort in mass production. All the maker can do is clean up his sand and vats and sample-test. Only a high-value-added assembly ($1,000 mike amp) could afford to check every box for high bass random-rumble. So you have generic switch transistors for low-price gear, well-made MAT or LM394 devices in better gear, and the best gear has some poor guy listening to the noise of every unit with a dummy input termination, and sending any oddballs back for reworking.
So yeah, for a BJT design optimized for 200Ω, 10uFd or 15uFd is not horrible. It may even be inaudible behind studio rumble and transistor excess noise. But I note even Mackie's more affordable models use 47uFd, so I assume you don't have to be hyper-fussy to notice a difference between "maybe enough" and "over-kill" input caps.
Another issue is that such caps are "always" electrolytics. (15uFd of film is not only costly, it is so big that it will catch all radio waves and buzz in the room. It can be done, but isn't the easy way out.) And electros distort. And over-size electros distort less (for the same load). And electro price does not rise much with capacitance. So if using electros, always go big.