This relates to a long-standing bug report about the previous release. Epydoc ignores the encoding specified in a Python source file and generates XHTML with either "ascii" or "iso-8859-1" encoding.
I've tested, and this is still true with 3.0. I'm not sure if it's fixable, or even whether you want to fix it. Let me know whether you want me to keep the Debian bug open.