So here it is the original PD patch, that comes as an example with PD and everything. I made just a couple of very MINOR adaptations to test the bug. All I do is resize the sample array to 1 minute, then I start recording an oscillator in it, and half a second later I start playing it at normal speed (100).

Get the patch here http://sites.google.com/site/porres/I07.phase.vocoder.bug.pd

click on the collored bang and just wait for it.

Now check at the location atom box how - at around 47500 (47 seconds and a half) - the whole thing gets ruined. I'm making a very big shift to 3 octaves below so it's clearer how it all gets screwed up and dropped even lower all of a sudden so you cannot even hear it anymore.

This is an extremely crazy behaviour which I have no explanation whatsoever, and keeps making me doubt about the very laws of physics themselves. By the way, I was recording other stuff in my current patch that I've been tweeking with over years and at around 47.5sec the same thing happens... my patch can loop around the recorded buffer, and same stuff happens when I read over it again... and also at diferent pitch shifts... but somehow it sounds "ok" at 0 shift, so I can see that the Phase Vocoder actually stops working somehow. Now it seems this has something to do with some drawback and limitation of some of the objects involved, such as [tabread4~] or something. But I just dont know them that well to grasp the reason why.

Thanks a lot.