Hey...
Would you like me to make the samples
available?
no, i understood. Have you tried to trim the soundfile exactly from
where the first sample of the actual information starts to the very last
sample of "useful" information? Well, if there is silence at the
beginning of the information (even only 10-20 ms) the algorithm could
duplicate these chunks as well since its "just information" for it (as
long as it doesn't use a gain-threshold or something).
Otherwise, i unfortunately don't know how these algo's work....
What about asking the author? since they are always happy for bugreports
and stuff :)
best,
sebastian.