@techreport{oai:ipsj.ixsq.nii.ac.jp:00216850, author = {高野, 哲朗 and 能勢, 隆 and 金垣, 葵 and 渡辺, 聡 and Tetsuro, Takano and Takashi, Nose and Aoi, Kanagaki and Satoshi, Watanabe}, issue = {4}, month = {Mar}, note = {健常話者の音声から構音障害話者の音声へと声質変換することにより,障害話者の声色を維持しつつ聞き取りやすい合成音声を生成する検討を行った.少量の障害話者データであっても鮮明な音声が得られる多数話者ボコーダを利用し,話速変換による本人性の向上や,音高拡張による抑揚の単調性の解消,ファインチューニングによる単語データのみの学習でその有効性を示した., In this study, we investigated the possibility of generating intelligible synthetic speech by converting the voice of a normal speaker to that of a dysarthric speaker while maintaining the tone of the speaker’s voice. Using a multi-speaker vocoder which can produce clear synthetic voice even with a small amount of impaired speaker data, we demonstrated the effectiveness of speech rate conversion to improve voice similarity, pitch augmentation to overcome monotonicity of intonation, and fine tuning to learn with word data.}, title = {健常音声からの声質変換と多数話者ボコーダによる構音障害話者の明瞭な音声合成の検討}, year = {2022} }