We aim to provide controls for emotion in synthetic speech. Many emotions are not displayed continuously in an otherwise emotional utterance; rather, the intensity varies with time. We show that an emotion recogniser is capable of producing a measure of emotion intensity via attention or saliency; this measure is appropriate to label utterances subsequently used to train a speech synthesiser. We evaluate novel and published means to do this showing that, whilst it is no longer state of the art for emotion recognition, attention is a good way to indicate emotion intensity for speech synthesis.
Cite as: Schnell, B., Garner, P.N. (2021) Improving Emotional TTS with an Emotion Intensity Input from Unsupervised Extraction. Proc. 11th ISCA Speech Synthesis Workshop (SSW 11), 60-65, doi: 10.21437/SSW.2021-11
@inproceedings{schnell21_ssw,
author={Bastian Schnell and Philip N. Garner},
title={{Improving Emotional TTS with an Emotion Intensity Input from Unsupervised Extraction}},
year=2021,
booktitle={Proc. 11th ISCA Speech Synthesis Workshop (SSW 11)},
pages={60--65},
doi={10.21437/SSW.2021-11}
}