The preferred Codec setting on the phones is G729a but an inbound call connects with G722 for both encode and decode, and an outbound call connects with G711a for both encode and decode.
Decode latency varies between 90 and 190 msec for the outbound/inbound calls respectively, but the round trip audio delay is measured roughly at 0.5 sec by one party saying stop to the other party counting one-and-two-and- ...
Unfortunately the Cisco SPA525G2 is reporting the R Factor, MOS-LQ and MOS-CQ fields as Not Available.
SPA series can only measure MOS stats in a CUCM environment.
You have all sorts of potential transcoding occuring, and if you haven't configured the SPA correctly (packet size needs to be changed to 20ms from the default 30ms for G.711) could be seeing some sort of issue with ptime mismatch depending on the SIP proxy.
If there is G.711 to G.722 transcoding occuring (since an inbound call from the Spark PSTN for example can't be native G.722) you could be getting audio issues as a result of the transcoding. Those Cisco phones don't seem to handle this sort of environment that well.
Who actually set this up? I would be configuring things correctly and ensuring they are set to G.711a and disabling G.722 for a start to see what impact that has.