The function sync_recognize() of the Speech API has consistently a shorter runtime, if I use a flac audio file with a sampling rate of 48 kHz in comparison to a sampling rate of 16 kHz.
How does it come that the larger file with more information to process, is analyzed faster?

And as a follow-up question: The docs say that a sampling rate of 16 kHz yields the best results. What is the reason for that?

(I already asked the question on Stack Overflow with some code included, but got no answer so far.)