So I just made a few recordings of me talking through Librem 5 Dogwood and Nokia N900. The receiving (and recording) end was Librem 5 Birch where the calls have been initiated:
https://dosowisko.net/l5-call-audio/call-rec-l5.flac
https://dosowisko.net/l5-call-audio/call-rec-n900.flac
And here are the exact same calls recorded directly on the devices I’ve been talking to:
https://dosowisko.net/l5-call-audio/call-rec-direct-l5.flac
https://dosowisko.net/l5-call-audio/call-rec-direct-n900.flac
Seems like we’re better at recording the room’s acoustics than N900 because of the stereo mics. The modem’s audio does sound a bit different between the devices. I wonder whether it’s the PulseAudio’s intelligibility enhancer’s output being slaughtered by the GSM network’s codec? 
When I get some time for that I’ll try to play with some knobs and check whether it can be improved further, but it seems good enough to me for now so I don’t think it’s the most urgent thing to do at the moment. Anyone’s free to play with it by themselves and report the results though 