EXOVOICES
Overexposure to screens has a negative impact on children's cognition, while audio stories can have a positive impact on their attentional and imaginative engagement. Virtual storytelling using synthesized voice can improve both engagement and imagination by addressing the child by name, for example. Current vocal synthesizers suffer from monotony and lack of expressiveness. EXOVOICES proposes a neural synthesis augmented with automatic linguistic annotations of the text (sentiment, semantic and structural analysis), which will generate a more diversified synthesized voice according to the passage of the story and the expressive tone of the text. Existing evaluation protocols are not adapted to reflect the monotonous and expressive character of a synthesized voice; EXOVOICES intends to contribute to the definition of the notion of expressivity by elaborating an experimental protocol to evaluate this notion on natural and synthesized voices.
The impact of listening to natural and synthetic stories on children's attention and imagination will be evaluated through cognitive experiments, which will constitute a new evaluation paradigm for voice synthesis. A better understanding of the mechanisms of attention fluctuation during reading and listening will have important implications for education and for the management of children with attention disorders.