A polyglot domain optimised text-to-speech system for railway station announcements

Announcements at railway stations are a major information source for passengers. In order to ensure high intelligibility, the traditional solution is to use recorded prompts with “slot filling” of variable data. If a data type (e.g. train name) changes new recordings have to be made. Even with careful design the quality of the system will gradually deteriorate due to change of the voice of the voice talent, speech rate, etc. Advances in corpus-based technology have allowed the introduction of text-to-speech solutions into this application domain. In this paper our solution for a flexible, single voice based polyglot system is described. It is currently implemented for Hungarian and English with plans underway for German. Hungary, being at the geographic center of Europe is at the crossroads of rail connections to more than 15 countries. The Hungarian system announces the Hungarian variant of station names while the English system shall read them in the official language of the country (e.g. Venice is ‘Velence’ in Hungarian and ‘Venezia’ in Italian). The system has been in operation at the largest passenger railway station of Hungary since June 2014 and has been installed for more than 60 other stations and stops.