It appears that any AAC audio exported from Adobe Media Encoder is off by about 10 ms / 575 samples / a quarter of a frame.
Apparently, this has been a thing for at least five years.
I came up with a fix for this through FFmpeg:
ffmpeg -i filename -itsoffset 00.01 -i filename -map 0:v -map 1:a -c copy newfilename
My question is…does it matter? Could anyone hear that? Is it worth worrying about??
We only found out about it because the mixer put our MP4 against his mix.
Take 30 seconds of black and every second add a single frame of bars and time. Then make two copies and offset one to make the audio early by a frame and then offset the other one by making it a frame late. $5 says you can’t tell the difference.
If you can, send me your address.
Then, enable sub frame positioner and repeat with half frame offsets. If you can, another $5.
Then repeat with 0.25 frame offsets.
Could be quite a lucrative side hustle for ya Ted.
Think of it this way: If the audio is in perfect sync at the start of a frame, before that frame is done playing the audio will be nearly one frame out of sync. A quarter frame here or there isn’t going to make a difference. And for the record, every flame I have ever used since day one, the audio that was output would be a fraction of a frame different when it was re-input. It’s not just media encoder. Because of this “drift” I am one of the few people left that still believes in the almighty 2 pop.
And for laughs, we once had an audio client who claimed he could tell when something was a tenth of a frame out of sync. He was, of course, as full of shit as he was of himself.
Send your audio friend an uncompressed QuickTime and have them lay it back against their final audio, and then have them send that back to you. Laugh at the results. Crack open a cold frosty one.
adding this for cross reference:
Also one thing to consider:
Sound travels a lot slower than light, so lets say you sit 5m away from your speakers.
thats 14ms of delay right there… 
(340m/s speed of sound at room
temp approximately).
measure the actual delay using a highspeed camera its fun. I dont know how
many people watch tv… things are something off by a LOT.
very interesting.
For US delivery I still cant deal with not having set full frames for spots, not even going to get started on surround mixes 
5s spot how many frames is it, in pal world thats 125 in the US at 29.97 thats… 149 or 150 frames? do you round up? down? that would all make it just slightly too long/short for 5s of audio either way, it cant ever be perfectly 5s , my brain breaks from this everytime I think about it.
but then 150frames at 29.97 is more than 5s of length
Yeah. It’s cracked out.
Correct - NTSC timecode runs slower than a clock on the wall.
That’s why drop frame timecode is a thing.
You’re missing the bigger point which is that a non integer frame rate remains as idiotic now as it’s always been. PAL was always easier to deal with and when Europe transitioned to hd it was equally simple.
29.97 dropframe is a hill you want to die on, fine. The sane remain unconvinced.
Right… you go with that.
Sigh.