Interspeech'2005 - Eurospeech

Lisbon, Portugal
September 4-8, 2005

Reconstruction of Polish Diacritics in a Text-to-Speech System

Artur Janicki (1), Piotr Herman (2)

1Warsaw University of Technology, Poland; (2) Fincom-MATERNA Communications Ltd., Poland

This paper describes an approach to reconstruction of the Polish diacritic signs, needed e.g. in a speech synthesis system. Some telecommunication services (for example SMS transmission in GSM) remove diacritics from the text. Without them the text is usually still understandable to a reader, but if a TTS system reads it, the speech becomes heavily distorted. In this paper we propose to use neural networks to reconstruct the Polish diacritics. Architecture of the proposed system is described, the process of training and testing is presented. At the end a real-life implementation is described. Usage of SMS-to-speech service increased by more than 30% after implementing the proposed system of reconstructing diacritics.

Full Paper

Bibliographic reference.  Janicki, Artur / Herman, Piotr (2005): "Reconstruction of Polish diacritics in a text-to-speech system", In INTERSPEECH-2005, 1489-1492.