Multilingual number expansion for TTS

Abstract

As an inherently language-dependent feature, the expansion of numbers to their full written-out forms can also be accomplished with language-independent code for multilingual TTS systems. In this paper we present a number expansion system that works with little data and language-independent code, while still able to expand numerics in dozens of languages. The paper also describes a way to determine the type of the number for correct expansion type, also with language-independent code, normalization from various scripts, and discusses issues brought up by conjugation of some morphologically rich languages.