The distance is euclidian IIRC, and those two things are equally spaced, so I assume there is some other precedence that determines what is returned. My feeling is that it returns the first index it finds, so whichever entry occupies the lowest number idx will be returned. In that case I would ask for 5 or something and do some extra filtering at the end. No matter what you do you will always have that problem unless you specify another set of criteria to search by.
As you can imagine it is fast. The searching is at audio rate and I haven’t been able to make it crash yet, however, I haven’t profiled or looked into how I can implement the searching in a better way. Like I said its mostly untested and I just wanted to see if it worked out on some small scale experiments. Theoretically you can retrieve a match in one vector of audio…I think…