Study on utterance training system for improving likability by using speech processing technology

2019 Fiscal Year Final Research Report

• Project/Area Number: 16H05899
• Research Category: Grant-in-Aid for Young Scientists (A)
• Allocation Type: Single-year Grants
• Research Field: Educational technology
• Research Institution: Meiji University (2019); University of Yamanashi (2016-2018)
• Principal Investigator: Morise Masanori 明治大学, 総合数理学部, 専任准教授 (60510013)
• Project Period (FY): 2016-04-01 – 2020-03-31
• Keywords: 教育工学 / 音声情報処理 / 声質変換

Outline of Final Research Achievements

We carried out a series of studies to develop a system to measure and improve the perceived likability of speech. The experiments showed that there was an appropriate range in speed of speech and duration of pauses. In pitch and timbre, we found the speech parameters related to the perceived likability. In this study, we utilize a vocoder technology to control the pitch of speech. In the timbre, we proposed a signal processing method to control the spectral centroid, which roughly corresponds to the brightness of the speech. A subjective evaluation using the proposed method was carried out. The result suggested that it was possible to improve the likability of the input speech. We implemented a prototype of a speech-likability measurement system that incorporates these technologies and confirmed that it works in a real-world environment.

Free Research Field

音声情報処理

Academic Significance and Societal Importance of the Research Achievements

本研究は，音声認識のようにテキスト情報を扱うものではなく，音声に対して知覚する好感度という，被験者に依存する感性情報を扱うという挑戦的なテーマである．現在の音声認識・合成技術の性能は，すでに人間と等価な水準に達しつつあるが，このような感性情報の計測・制御に関しては，被験者に対する依存性もあるため研究事例そのものが相対的に少ない状況にある．一方，発話訓練には社会的なニーズがあり，コミュニケーションの支援技術は今後需要が増加することが期待される．本研究で得られた成果は学術的にも価値があり社会的にも還元しやすく，今後類似した研究を進める指針として役立つと考えられる．