Based upon thorrild's bug report and surfwhammy's examples of the embedded XML at the end of the .prox files, I think I
may have found the bug. I am looking at byte offset 0x0B8318C9 in "Bassoon Duo Expansion.prox" . This is the sample set for {some type of articulation, not sure which yet} at dynamics of either p or mp. The suspect line of XML reads
<sample codec="1" data-start="123412992" data-size="1079344" sample-channels="2" sample-frames="269836" frame-size="4" tone-marker="0" release-marker="0" loop-begin="88712" loop-end="190263" nso-key="0" base-pitch="Bb1" lower-pitch="A1" upper-pitch="C2"/>
and the discrepancy is that all the other samplesets in the file use "B1" as the base pitch for that range, not "Bb1".
However, I changed it to
<sample codec="1" data-start="123412992" data-size="1079344" sample-channels="2" sample-frames="269836" frame-size="4" tone-marker="0" release-marker="0" loop-begin="88712" loop-end="190263" nso-key="0" base-pitch="B
1" lower-pitch="A1" upper-pitch="C2"/>
...but that didn't change the behavior
. Anyhow, my guess at the semantics of the base-pitch attribute would,
if this were the cause, have made the affected pitches sound a half-step
low instead of a half-step high. (These attributes look like they mean "if the pitch I'm asking you to render is between lower-pitch and upper-pitch, then go get the sample at data-start for data-size, interpret it as representing a snippet of base-pitch, and transpose it accordingly if necessary." But this doesn't seem to be quite right, else my change would have had an audible effect, possibly raising the problem notes yet a further half-step.)
You may imagine how delighted I would have been to be able to present a fix along with the diagnosis! But something else is going on. It may be that the samples themselves are wrong. I'm not going to start mucking about trying to find a different block of sample data. Come to think, the change of the base pitch to Bb1 may have been someone's unsuccessful attempt to correct the problem at the metadata level.
Thanks,