Text-to-speech systems are normally not suited for name pronunciation in, for example, an automatized reverse directory service. A new project has been started to alleviate these problems. First names, surnames and street names corpora from the Greater Stockholm telephone directory has been studied. The surnames corpus has been tagged as to language origin. Both rule based methods and an artificial neural net classifier has been tried to predict this information. Specialized rule components are currently being written to predict the pronunciation for names of different origin.
Bibliographic reference. Carlson, Rolf / Granström, Björn / Lindstrom, Anders (1989): "Predicting name pronunciation for a reverse directory service", In EUROSPEECH-1989, 1113-1116.