ISCA Workshop on Multilingual Speech and Language Processing (MULTILING 2006)

Center for Language and Speech Technology, Stellenbosch University, Stellenbosch, South Africa
April 9-11, 2006

Investigating Prosodic Modifications for Polyglot Text-to-Speech Synthesis

Péter Olaszi, Tina Burrows, Kate Knill

Speech Technology Group, Cambridge Research Laboratory, Toshiba Research Europe Limited, Cambridge, UK

This paper investigates the need for applying English prosody when synthesising English portions of mixed English/German texts using a German-based polyglot text-to-speech (TTS) synthesis system. The polyglot system is based on a monolingual German TTS system, which uses a phone mapping from English to German to synthesise English texts. Two systems with varying degrees of assimilation to English are compared, one in which prosody is derived from the German monolingual system, and one in which the prosody is derived from an equivalent English monolingual system. The naturalness of the different prosody approaches and overall intelligibility and acceptability of the polyglot systems is assessed by native bi-lingual speakers of both English and German, on German texts containing varying lengths of English inclusions, and on complete texts in English. The results show that both German and English subjects preferred English prosody for longer English inclusions or complete English texts, but had no preference for short inclusions.

Full Paper

Bibliographic reference.  Olaszi, Péter / Burrows, Tina / Knill, Kate (2006): "Investigating prosodic modifications for polyglot text-to-speech synthesis", In MULTILING-2006, paper 017.