In a nutshell... the TransMetaphone package implements
the DoubleMetaphone algorithm for 10 different languages
where keys are generated in IPA symbols (more than 2
keys may in fact be generated).  The usefulness of which
is that you can then compare words (presumably phonetically
transcribed proper nouns) across language and script
boundaries.  Applied to text retrieval, you can can generate
a key in one language and then search for it any other
supported language.

Such is the goal, an extended description is provided in
the doc/index.html file.  See also the doc/KeyTable.html
file, which is not perfect but not too far off either.