Voice Puppetry with FastPitch

Affective speech synthesis is an active research area, but recent approaches usually lack the full, fine-grained controllability to produce utterances with any exact affect intended by the user. We propose a puppetry tool based on FastPitch to help model output convey any required suprasegmental meanings. Users can choose any trained FastPitch model, and which features should be mimicked, making the approach fine-grained and language-independent.