Word-to-phoneme translation

[Asger Alstrup Nielsen]

> (3) Then again, a to-phenome algorithm would also work.  The only
> question is getting enough and the right kind of test data.  Does anyone
> know of a publicly available pronecation dictionary?

The guy that did some work on pronounciation did some work on word-to-phoneme
translation too.  Check out

http://ilk.kub.nl/~antalb/pubs.html

However, I suspect his work is in Dutch only.

There is a big open source project that does artificial speech.  I forget the
name.  Embrola maybe?

They have code that does very good word-to-phoneme translation for many
languages.  Try to hunt for speech on freshmeat.  Using that code would
arguably be the easiest solution.

> My only question is how fast would it be compared to my current method?
> (Sorry I am really AI illiterate)  In particular how fast would an AI
> based to phoneme conversion be compared to a hand tuned one.  Also which
> one would you think would be better the first or the second?

A neural net takes time proportional to the number of connections in it.  So
it's difficult to say in general.  If the net is big, it'll be slow.  If it's
small, it'll be fast.  There is no easy way to know how big a net should be
before you try it out.  You (or the computer) have to do a lot of trial'n'error
to get the best network topology.
But the advantage of a neural net is that it would not be language specific. 
Given a different training set, the same network could arguably be used for any
language you have data for.

Greets,

Asger