The IVO Software Blizzard Challenge 2009 Entry: Improving IVONA Text-To-Speech

• Main Authors: Michal Kaszczuk, Lukasz Osowski
• Other Authors: The Pennsylvania State University CiteSeerX Archives
• Format: Text
• Language: English
• Subjects: Index Terms; IVONA; IVO Software; Speech Synthesis; Blizzard Challenge; Arctic
• Online Access: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.178.9954; http://www.festvox.org/blizzard/bc2009/ivona_Blizzard2009.pdf

This paper describes a special version of IVONA Text-To-Speech for a GB English voice designed and developed by IVO Software for The Blizzard Challenge 2009. The architecture of this system is based on an improved IVONA Text-To-Speech originally developed for previous challenges- Blizzard Challenge 2006[1] and Blizzard Challenge 2007[2]. This year we decided to build two GB English systems (using the full database and the arctic subset) and complete four challenge tasks EH1, EH2, ES2 and ES3. The system used for completing tasks E21 and ES3 as well as for task EH1 was built on the full ’roger ’ database. Hence we show a basic overview of the IVONA Text-To-Speech architecture. Then we focus on methodology and problems which we experienced during development of our GB English voice from the ’roger ’ database provided by CSTR 1. We also present a short analysis of the Blizzard Challenge 2009 results and future plans for development of IVONA Text-To-Speech.