Analysis and Assessment of Controllability of an Expressive Deep Learning-Based TTS System

In this paper, we study the controllability of an Expressive TTS system trained on a dataset for continuous control. The is Blizzard 2013 based audiobooks read by female speaker containing great variability in styles and expressiveness. Controllability evaluated with both objective subjective experiment. assessment measure correlation between acoustic features dimensions latent space representing perceptual experiment which users are shown interface Controllable asked to retrieve synthetic utterance whose expressiveness subjectively corresponds that reference utterance.