Bonjour!
Could it be that the attribute slice.hop indicates the number of hops instead of the size (as the documentation states)? Maybe my idea how onset segmentation with pipo.onseg works is wrong, but the number of segments found increases with smaller ratios of slice.size to slice.hop, i.e. with increasing hop size (or rather number of hops?), e.g. with a slice.size of 512, more segments are found with slice.hop 256 than with slice.hop 128.
The semantics of slice.hop put aside, how would one calibrate these two attributes according to the musical content to be analyzed, for example soft attacks and tied notes vs. percussive attacks, density per second etc.
Some hints would help me a lot to improve the usage of this great tool.