Polyglot text to speech synthesis

The conversion of mixed-lingual text, i. e., texts that contain inclusions of multiple other languages in forms of phrases, words, or even parts of words, into artificial speech signals poses several problems to doday`s text-to-speech syntheses systems. This thesis describes a new polyglot text-to-speech syntheses system that presents a solution to these problems. Concretely, new approaches to mixed-lingual text analysis, to mixed-lingual phonological processing, and to polyglot prosody control are presented in detail. This system, as intended for the multilingual setting of Switzerland, is currently able to analyze text of any mixture of English, French, German, and Italian, and to generate artificial, polyglot speech signals in these languages with polyglot prosody consisting of French and German parts. In order to restrict this polyglot text-to-speech systhesis system to a certain multilingual setting, strong emphasis is put on a modular system architectue. This architecture allows the construction of a polyglot synthesis system by combining independent, monolingual resources.