[Toxaris]

I usually work from a Word file, often result of an OCR scan process. Then I clean the whole document up and fix the OCR errors with my add-in. Then I directly export from Word to an ePUB. Often I use a standard stylesheet which is automatically inserted into the ePUB (and linked of course) when I export it as an ePUB from Word. As a final touch I load the ePUB into Sigil for some work if needed.

Needless to say is that the XHTML in the ePUB created in this way does NOT contain any of the Word HTML garbage and that it can handle things like tables, foot- and endnotes, images, equations and so on. If you want, you can even generate a stylesheet based on the styles used in the Word document.