SANE-TTS: Stable And Natural End-to-End Multilingual Text-to-Speech. (arXiv:2206.12132v1 [eess.AS])

In this paper, we present SANE-TTS, a stable and natural end-to-end
multilingual TTS model. By the difficulty of obtaining multilingual corpus for
given speaker, training multilingual TTS model with monolingual corpora is
unavoidable. We introduce speaker regularization loss that improves speech
naturalness during cross-lingual synthesis as well as domain adversarial
training, which is applied in other multilingual TTS models. Furthermore, by
adding speaker regularization loss, replacing speaker embedding with zero
vector in duration predictor stabilizes cross-lingual inference. With this
replacement, …

arxiv natural speech text text-to-speech tts