This paper describes the system with which we took part in the Blizzard Challenge for the first time. It describes how we created our own annotation from scratch and introduces the various system components, in particular the back-end. Results show that our system achieves excellent intelligibility, in particular with less data (e.g. the lowest word error rate for voice C 1), and reasonable naturalness (MOS of 3.0, 2.9 and 2.9 for voice A, B and C, respectively). They also reveal that our simple approach of randomly selecting sentences for voice C worked well.
Bibliographic reference. Buchholz, Sabine / Braunschweiler, Norbert / Morita, Masahiro / Webster, Gabe (2007): "The Toshiba entry for the 2007 Blizzard Challenge", In BLZ3-2007, paper 014.