Just thought sod it and connected a analogue mic module without dc blocking caps.
3.3 -> Module VCC, GND -> Module GND
Out -> Mic-P, GND -> Mic-N
Works fine the mems is a bit quiet and the electret with the MAX9814 is mighty as it usually is.
There was DC buzz but far less than my lazy approach expected as thought it would be horrendous but ok just to test.
I will get some .1uf ceramics on Mic-N & Mic-P and see how its goes.
Also while I am at it a 4.7uF DC blockers on the stereo out is prob worth a try also, so will come back on that.
[edit]
I would say these are definitely the best to go for as even though electret rather than mems the Max9814 gives far more amplification.
There is a level of dc buzz that I think is endemic to the Soc board as on high gain with the codec you just amplify this base buzz with you signal.
So the higher the input signal the less you need to amplify.
Yeah blocking caps help a little but going to play with values a little bit more as its guess work to the impedance of the mic module.
Also because I had one I put an audio cap 12000uF across 5v & gnd which did clean the line a little but minimal.
Prob smaller 
might help and not lessen the effect but debatable if worth the hassle or space.
The mems analogue modules due to the large codec gain needed produce quite a lot of noise. Its still completely fine for recognition but for voice recording you wouldnāt be happy.
In fact even the max9814 you probably wouldnāt be happy but in comparison to the mems the low level of background noise is only a problem if you where thinking of creating a wifi mic or something and expecting studio quality.
My application of a voiceAI then more than happy with the sensitivity of the max9814 you need to put 3.3v on the gain pin as the top gain by leaving it floating as usual with these modules is noisy.
Dunno why they are that way as the top gain is pure over optimism whilst -10db makes a world of difference.
The low gain by applying gnd to the gain pin doesnāt seem to be much different but the lower signal it produces means more codec amplification and the board noise starts to creep in.
You can listen here to the electret with a codec gain of 13.5db with a single mic
https://drive.google.com/file/d/1f4IUcTLdSJDCfdRAsQuA4JSWxw39670t/view?usp=sharing
Also my computer is pretty noisy so here is a recording where through the recording it goes to sleep and you can hear in a much quieter room a truer reflection of quality.
https://drive.google.com/file/d/1goGPNXlUE8P-R2T3tHrhjXFQwtWENz0h/view?usp=sharing
