Unparse-text() string contains ascii chars 29, 30 and

I'm trying to process some data that's one long string delimited using
ascii characters 29, 30 and 31 (which are apparently group, record and
unit 'separator characters').

I can get access to the string using unparsed-text(), but when I
attempt to process the string using any of the function eg:

tokenize($str, '')

or

substring-before($str, '')

...the XML parser complains that these aren't legal XML characters
(when the stylesheet itself is parsed).

Is there any way around this?  I can't see how I can process the
string in XSLT without using the characters themselves.

The two alternative's I can see are to use an XMLFilter to turn it
into XML using Java, or to go back to the source to get them to export
their data in a less archaic way...