The CSTR entry to the Blizzard Challenge 2016
Thomas Wayne Merritt, Srikanth Ronanki, Zhizheng Wu, Oliver Watts · 2016
Similar to 2016 and 2017 Blizzard Challenge, the task for this year is to train on expressively-read children’s story-books, and to synthesise speech in the same domain. This give us an opportunity to investigate the effectiveness of several techniques we have developed when applied to expressive and prosodically varied audiobook data. This paper describes the text-to-speech system entered by The Centre for Speech Technology Research into the 2018 Blizzard Challenge. The system is a hybrid synthesis system where a halfphone unit selection synthesiser is driven by the output of a neural network based acoustic and duration model. We adopt the same neural network based models used in our last year entry with a different unit selection component. We discuss the performance of our system by reporting the results from formal listening tests provided by the challenge.