Very nice, I hope this comes in handy. I didn't want to turn this into a TLDR kind've post, but there are limitations to these different methods. If you have a continuously varying delay (i.e. every sample), for instance, the recursive algorithms probably aren't for you, maybe the "smooth" method instead. Something like the Hermite might be, but you'd need to make sure you get four data points, that is, two samples before and two samples after the spot where you want to get that fractional delay. I guess the point of it all is that having algorithms which don't kill your top end as much as the linear method allow you the freedom to choose how exactly you want to filter the results.