Investigating Flex 6700 total latency
Inspired by Andy KU7T excellent write-up I wanted to look at my  own flex radio 6700 setup and measure the TOTAL latency at around 3500km. 
The problem: 
Whilst using the flex maestro remotely, there is a perceivable high audio latency which makes it extremely difficult to use for high rate contests. KU7T in his tests showed that the flex takes a mimimum of  48ms to process an RF signal and output it to the speakers at the front of the radio, so the minimum latency you can expect is around 50ms...
The question: 
So I wanted to find out what is the Total Latency from Signal to Ear at 3500km for my 6700.
Inspired by KU7T I measured the following setup :
6700--- ROUTER---ADSL------3500km----------FIBRE----- ROUTER---PC or Ipad or Maestro [no wifi anywhere]
To measure the latency I used a  win 10 PC (@3500km from radio) with  a behring uca-202 USB sound card attached to it. Any sound card with stereo line-in sound card should work.
The UCA202 uses RCA as inputs for L and R channels. I used 2x RCA to mono phono cables. One for the REFERENCE signal (real time) [L] and one for the test signal [R].
The test signals were one of following:
a) Maestro model A (running v 3.1.12), connected via ethernet, using audio from headphones output 
b) Ipad Pro 2020, running smartSDR app (V3.9.12) with a VA-UC006 USB-C dock, ethernet and SoundBlaster SB1920 sound card with audio out
c) RemAud (by DF3CB) v2.1 encoding at PCM 22Khz, 16bit stereo (10ms buffer). I used my PCs integrated sound card and connected the RCA to phono cable from the  headphone output of the PC to the R INPUT  channel  of the UCA202.
Variables and other settings:
Filter sharpness (maestro menu->Phone/CW->filter options): Settings 1-4 (1=low latency)
Ipad SmartSDR app was set to NO (disabled) Large audio buffer (Settings->extended Settings) 
For real time reference I used a KX3 (@3500km from flex) with a 5m wire as an antenna and tuned onto any signal I could hear on 40m.
I then tuned the flex at exactly the same frequency and using audacity I captured BOTH signals as described by KU7T and calculated the latency.
Tip: In audacity select:  show Length and end of selection at the bottom of the screen and time format hh:mm:ss + milliseconds
I measured differences on 2-3 different signal times to ensure repeatability. 
Network latency was obtained using the ping command. This is Round Trip Time. In my case in was 95ms ie approx 48 ms each way.
All values in ms 
Ipad Pro 
Level 1 filtering 
Level 2 filtering  
Level 3 filtering  
Level 4 filtering  
Results show an unacceptably high latency for maestro of 540ms (588-48). Using an Ipad Pro seems to have the best result with a total latency of 88ms. The non network latency for the ipad is hence 40ms which is comparable with what KU7T has published for the time it takes the 6600M to output audio from the speaker jack.  This is an impressive 6.7x better than the best result achieved for the maestro. RemAud gives an alternative acceptable latency of 155ms. In his tests, KU7T showed that latency can be as low as 98ms using this technique.
For this experiment I made the assumption that the KX3 has no processing latency to output the signal.  
 To achieve lowest total latency it seems that best option is to use an ipad pro with smartSDR, large audio buffer disabled and flex filter options setting set to Low latency. My maestro is now demoted to smart control!
I plan to repeat the experiment to improve repeatability of the values. 
Many thanks to KU7T for sharing the methodology to the world and inspiring more experimentation. 
UPDATE 22/02/2021:
Flex Helpdesk acknowledge that this is a known issue and is to do with Windows Audio latency issues beyond their control but are looking at possible solutions!