Inspired by MarkT's insisting for numbers, I did some additional evaluations.
Teensy 4.1 with audio shield supplied from and connected to my PC via USB.
Setup open without shielding. MCP602 in buffer.
Switched over from peak.read() to rms.read() which leads to nicer numbers for same physics.
Better to compare to known numbers.
Comparison of a line input, which is directly connected to analogue GND to one, which is connected to the buffer. The buffer input is connected to GND.
For my Humbucker Guitar a line sensivity of 1.33Vpp without gain is good.
As the noise RMS value shall be compared to a sine signal with maximum amplitude, the RMS number is reduced by 1/sqrt(2) = 3.1dB to form a RMS Signal-to-Noise-Ratio.
So all in all the high impedance input buffer with gain=1 lead to a degradation of unweighted Signal-to-Noise-Ratio from 77,8 to 76,7dB for sensivity of 1.33Vpp.
It is astonishing, that lower line sensivity does only show better SNR, if the buffer with higher gain=4 is used.
These numbers are quiet nice in comparison to top analogue UKW stereo receivers, which could achieve about 70dB for stereo.
http://www.hifi-forum.de/viewthread-84-783.html
Wrapping the whole setup with aluminium foil for shielding did not improve SNR.
Datasheet of SGTL5000
https://www.nxp.com/docs/en/data-sheet/SGTL5000.pdf says, that SNR of 85dB can be achieved.
As far as I understand SNR=76,7dB corresponds to "Effective Number of Bits" ENOB =13.2 bits
Has somebody done similar investigations about SNR of the audio shield?