Cost-efficient cross-lingual adaptation of a speech recognition system

Abstract

In this paper, we describe a methodology that proved to be successful and cost-efficient for porting an existing speech recognition system to other languages. Our initial aim was to make a system called MyVoice developed for handicapped persons in Czechia, available also to users that speak other languages. The experimental results show that the proposed method can not only be used with languages with similar origins (in this case Czech and Slovak), but also with languages that belong to very different branches of the Indo-European language family (such as Czech and Spanish), obtaining in both cases accuracy rates above 70% with a 149k words lexicon.