Right: Here's a thread with my take on Yannelgate. I have now ripped the audio from the ambiguous token, and the proposed original "Laurel" from Vocabulary.com 1/
I put both into Praat, with Pitch Settings 50-200Hz, and Formant Settings to 5 formants under 5000Hz (both settings suitable for adult male voice) 2/
Here's the original "Laurel", showing a rise-fall pitch contour and 5 formants, with F1 and F2 very close together. 3/ 
And here's the ambiguous "Yanny/Laurel" clip, showing a very similar pitch contour, but with no energy above ~6.5kHz plus only one formant in place of the F1-F2 complex of the original. 4/
FYI blue is pitch, red dots show formants (F1, F2, F3). My cheap take on this would be that one very strong cue to the "r" sound in English is a dramatically lowered F3 - see how that dips down in the original "Laurel" around the middle of the word. 5/
So, if the F1-F2 in "Laurel" is - probably due to a slightly ropey recording mic - fused into one reasonably faint F1, then for the listener there's a risk that the original F3 is actually heard as the F2. 6/
Hence in "Yanny/Laurel", the perceived F2 is the dipping formant, which in combination with the much lower F1 now more strongly corresponds to the sounds at the start of "yanny" (maybe sounding like an /i/ going into an /ae/ and even the nasal /n/). 7/
Also there may be some additional effects of there being no signal >6.5kHz in "Yanny/Laurel"... So, I tried it out by making my own version of the ambiguous sound: applied some pre-emphasis to all sounds above 600Hz and added the lopass. Here's a pic: 8/
And finally, here's an mp4 of my "Yanny/Laurel", created from the original "Laurel" file as described in my previous tweet. I'm incapable of perceiving any Yanniness even in the Reddit version by now... 9/9
