Emotion detection from speech using high-accuracy emotional speech recognition

2021 Fiscal Year Final Research Report

• Project/Area Number: 19K12014
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 61010:Perceptual information processing-related
• Research Institution: Yamagata University
• Principal Investigator: Kosaka Tetsuo 山形大学, 大学院理工学研究科, 教授 (50359569)
• Project Period (FY): 2019-04-01 – 2022-03-31
• Keywords: 感情認識 / 音声対話システム / 深層学習 / 言語情報 / 非言語情報 / 音声認識

Outline of Final Research Achievements

Research on speech emotion recognition has made great progress by using deep learning. However, the methods using non-verbal information contained in speech were the mainstream in the past, and the study of the methods using linguistic information were insufficient. In this project, in order to use linguistic information, emotion recognition was performed using the texts obtained by speech recognition. In addition, the above recognition results were fused with the results by non-verbal information to realize highly accurate emotion recognition. Finally, the recognition rate of 82.75% was obtained by recognizing four emotion categories.

Free Research Field

音声情報処理

Academic Significance and Societal Importance of the Research Achievements

音声を対象とした感情認識の研究において，これまでも言語情報と非言語情報を融合した感情認識の検討は様々行われてきたが，最大の問題点は音声認識の性能の低さであった．この性能が低いと正しい言語情報が利用できず，これが融合法のネックとなっていた．本研究ではその問題を克服し，高精度の感情認識を実現した．
音声による感情認識の精度が向上すれば，人間同士の対話に近い人対機械の対話が可能になり，ロボットの活用の幅が広がると考えられる．さらに本技術はコールセンターやメン
タルヘルスケアなど音声を使う様々な分野で利用可能と考えられる．