Expression team participated to the Blizzard speech synthesis challenge in 2018. The main goal was to synthesize audio books for children as for the 2017 edition. Here are some samples of our system output:
After multiple corrections to the system, the output has been improved as the following examples show:
In our case, the annotation process is fully automatic. Manually checking the phone segmentation should improve the results. Still, some errors are due to the lack of some phones in the corpus and the high level of expressiveness which is still difficult to control.