The Bell Labs German text-to-speech system

✍ Scribed by Bernd Möbius

• Publisher: Elsevier Science
• Year: 1999
• Tongue: English
• Weight: 711 KB
• Volume: 13
• Category: Article
• ISSN: 0885-2308
• DOI: 10.1006/csla.1999.0127

✦ Synopsis

The Bell Labs multilingual text-to-speech system can be characterized as consisting of a set of language-independent modules. Any language-specific information is represented in, and at run-time retrieved from, precompiled tables, models and finite-state transducers. In this paper we present a detailed description of the German version of the Bell Labs text-to-speech system. We will first discuss aspects of text analysis and our solutions to the problems they pose. Some of these problems, such as the expansion of numbers and abbreviations, and proper name pronunciation, occur in many languages while others, such as productive compounding, are specific to German and several related languages. We will then report on the construction of models for segmental duration and intonation. Finally, we will explain the design and structure of the acoustic inventory for concatenative synthesis and the criteria and procedures that were used to build it.