I like utalu. I suppose you could pronounce it either OO-TALL-OO or YOO-TUH-LOOH or OO-TUH-LOO (which I prefer, because it rolls off the tongue better and sounds really playful).
I think it will have a lot to do with the knowledge of the speaker. I saw lots of nihongo looking stuff, plenty of swahili and polynesian looking things too.
Edit: Though it all starts to blur together after the first 10,000 :-)
Definitely agree with you, it's contextual to each person. I'm Japanese, Japanese stuff stands out to me. Out of curiosity what do swahili and polynesian ones sound like?
EDIT: why can't I reply to you joshu? That sounds interesting, would be curious to see the results.
I've been working on language modeling. I built a word generator which makes medical-sounding names. I wonder if you could rank japaneseness, polynesianness, medicalness, etc.
If there's nothing here that is satisfactory, then it's quite easy to generate random pronounceable words, just alternate between vowels and consonants. As an example: https://github.com/Stuk/generate-words
I started out with something similar to this, but randomly playing combinations created a lot of garbage relative of things that sounded pronounceable. A lot of letters are simply unpopular/bad (looking at you Q, J, Z, X) in most combinations (often the exception is the end).
Also, a clever use of this data set would be to run it against the Google Search API, and order them by the number of results. Theoretically, more results = better word. Or, less results means easier SEO.