Recent Advances in Multilingual Text-to-speech Synthesis

In this paper we will discuss recent advances in multilingual text-to-speech (TTS) synthesis research at AT&T Bell Laboratories. The TTS system developed at AT&T Bell Laboratories generates synthetic speech by concatenating segments of natural speech. The architecture of the system is designed as a modular pipeline where each module handles one particular step in the process of converting text into speech. Besides conceptual and computational advantages, the modular structure has been instrumental in our effort toward a TTS system for multiple languages. This system will ultimately consist of a single set of modules, and any language-specific information will be represented in tables. In describing the TTS system, we will concentrate on the multilingual aspect, with a bias toward the German language.