2026-04-19T21:53:18Z https://ipsj.ixsq.nii.ac.jp/oai

oai:ipsj.ixsq.nii.ac.jp:00234657 2025-01-19T09:44:09Z 1164:5064:11558:11626

An experimental study of accent embedding for text to accented speech synthesis An experimental study of accent embedding for text to accented speech synthesis Hewei, Zhang Daisuke, Saito Nobuaki, Minematsu Hewei, Zhang Daisuke, Saito Nobuaki, Minematsu ポスターセッション2 In Text-to-Speech (TTS), End-to-End models had been introduced, which takes text as input and audio as output. This makes it hard to unsupervised control the style, especially accent, which consists of many kinds of acoustic features. We proposed a Phonetic Posterior-Gram-based unsupervised Accent Embedding Extraction model. Experiments showed the ability, robustness to different accent level of training dataset and deeper potential of the model to extract accent features from given utterance. In Text-to-Speech (TTS), End-to-End models had been introduced, which takes text as input and audio as output. This makes it hard to unsupervised control the style, especially accent, which consists of many kinds of acoustic features. We proposed a Phonetic Posterior-Gram-based unsupervised Accent Embedding Extraction model. Experiments showed the ability, robustness to different accent level of training dataset and deeper potential of the model to extract accent features from given utterance. technical report 情報処理学会 2024-06-07 application/pdf 研究報告音楽情報科学（MUS） 45 2024-MUS-140 1 5 2188-8752 AN10438388 https://ipsj.ixsq.nii.ac.jp/record/234657/files/IPSJ-MUS24140045.pdf eng