Is the voice contained as pain text with timer information and sent to Google translate and the translation displayed in real time, or are all Translations embedded in the YouTube video.
A very interesting question. Unfortunately, we can't say for sure - unless there are any Google insiders reading this.
Obviously, it does not make sense for Google to translate each & every time. So, I would imagine that they translate once and cache the result for future usage.
When they translate, we can only speculate:
- on upload?
- on first request?
- on Nth request (when they see that it is worthwhile)?
I imagine that they have an algorithm which they won't reveal, and are probably contently tweaking. At the end of the day, they want to balance good service (seeming real time translation) with cost, whether in servers, CPU cycles, electricity or programming effort.
Something getting a lot of views probably gets translated into the top 10 or so languages very quickly, where it might take a while for them to cache the Urdu version of my Welsh Vlog.
Tl;dr - they only ones who know are Google & they ain't telling, but we can make a few common sense guesses