Investigation of Japanese PnG BERT Language Model in Text-to-Speech Synthesis for Pitch Accent Language

End-to-end text-to-speech synthesis (TTS) can generate highly natural synthetic speech from raw text. However, rendering the correct pitch accents is still a challenging problem for end-to-end TTS. To tackle challenge of accent in Japanese TTS, we adopt PnG~BERT, self-supervised pretrained model character and phoneme domain We investigate effects features captured by PnG~BERT on TTS modifying fine-tuning condition to determine conditions helpful inferring accents. manipulate content being text-oriented speech-oriented changing number fine-tuned layers during In addition, teach information with tone prediction as an additional downstream task. Our experimental results show that pretraining contain accent, outperforms baseline Tacotron correctness listening test.