a1studmuffin wrote:On the topic of timestretching, I've always wondered why there isn't a program out there that has the option of a super super high quality time stretch, and what it does is, instead of just running the timestretch algorithm once with a set of parameters, it runs it 1000 or 2000 times with slightly different parameters each time, then mixes the results together. The idea behind that being that each timestretched result will have a certain amount of "error" and will be an approximation of the correct solution, and by spreading different error across each timestretch result + merging them together, you'd effectively reduce that error to a negligable amount. No idea if it would work. It'd be incredibly slow (ie. an offline process), just for cases where you wanted super high quality over performance. Maybe there's an error in my logic somewhere...

there is an error in your logic.

it's like saying that if you take 1000 explanations of september 11 and combine them together, you'll get the best approximation of the truth. your combined explanation though, involves ufos, giant lizards, godzilla, unicellular organisms bred by the cia + fbi + pakistan's isi, explosions, no explosions, and the number of planes definitely being both zero and four.

if you smash the results together, all the errors will accumulate: if you assume that a has an error at position 4 and b has an error at position 8, then a + b gives you one error at position 4 and one at position 8. iterate x1000 and you've got a pretty rubbish sounding sample, probably.