We have presented automatic (and near automatic) processes for building letter to sound rules systems from lists of entries and their pronunciations. We have successfully built LTS models for four different languages and feel confident this process will work for many other languages. As well as quoting results from held out data from the words lists used for training, we also present results of applying one model to unknown words from news text.
This method is fully implemented and documented and distributed with the Festival Speech Synthesis System [2], or a PERL implementation is available from http://www.cs.cmu.edu/ lenzo/t2p.