Development of speech articulation recovery support system based on a phonological / individual separation model

2020 Fiscal Year Final Research Report

• Project/Area Number: 17K01568
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Rehabilitation science/Welfare engineering
• Research Institution: Kumamoto University
• Principal Investigator: UEDA YUICHI 熊本大学, 大学院先端科学研究部(工), 教授 (00141961)
• Co-Investigator(Kenkyū-buntansha): 坂田 聡 熊本大学, 大学院先端科学研究部(工), 助教 (80336205)
• Project Period (FY): 2017-04-01 – 2021-03-31
• Keywords: ホルマント周波数 / 音声合成 / 障害者音声 / 個人性 / 音韻性 / 発話訓練

Outline of Final Research Achievements

We proposed a phonological / individual separation model based on the formant restoration function. As a result of verification of the personality characteristics included in the restoration function, it is possible to apply it to speaker identification from the correlation between the function shape and the speaker. In speech synthesis, by using a parallel type instead of the cascade formant synthesizer, it is possible to control the formant levels in the restored speech, which can contribute to the clarification of a characteristics of the impaired speech (muffled voice). Furthermore, we proposed a different speaker’s voice conversion algorithm via a standardized space as a phoneme conversion based on the hv-articulation space, constructed a PC-based real-time system, and confirmed its effect. For more practical use, it is necessary to verify the construction method of the normalization algorithm in the standardized space.

Free Research Field

音声情報工学

Academic Significance and Societal Importance of the Research Achievements

ホルマント周波数に含まれる音韻性と話者性（話者の個人性）を分離するための新たな手法であり、ホルマント分析・統計過程で生成されるホルマント復元関数に着目することで、ホ音声合成における話者変換、音声変換、話者識別・照合などの様々な音声応用システムの基本特徴量として利用できると考えられる。また、発話障碍者の劣化音声の健常化へ利用することで、障碍者・健常者間のオーラルコミュニケ―ションに寄与できるが、そのためには、障碍毎に異なると予想される健常化アルゴリズムの導入が課題として残されている。