Development of a noise-robust scream detection system using deep learning

2023 Fiscal Year Final Research Report

• Project/Area Number: 19K04935
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 25020:Safety engineering-related
• Research Institution: Osaka Electro-Communication University
• Principal Investigator: Hayasaka Noboru 大阪電気通信大学, 情報通信工学部, 教授 (50554573)
• Project Period (FY): 2019-04-01 – 2024-03-31
• Keywords: 悲鳴検出 / 悲鳴強調 / 深層学習

Outline of Final Research Achievements

The purpose of this research is to develop and disseminate a scream detection system that can be used in noisy environments. First, we proposed a scream enhancement method using deep learning to achieve high detection performance even in noisy environments. Through simulations assuming noisy environments, we confirmed that high scream enhancement effects and scream detection performance could be obtained. Next, to enable real-time operation on small PCs, we worked on reducing the computational cost. By utilizing the strong periodicity of screams, we succeeded in achieving the same or better scream enhancement effect with 1/19th of the number of parameters compared to conventional methods. Finally, in order to facilitate the discrimination between scream-like sounds and screams, we proposed a new method that applies the scream enhancement process described above two times, and confirmed its effectiveness.

Free Research Field

音声信号処理

Academic Significance and Societal Importance of the Research Achievements

本研究により，実環境で使用可能な新たな防犯システムの提供を可能にした．これにより，防犯カメラが設置できないプライバシーに配慮する必要がある場面における安全性が大幅に向上する．他にも，演算コストの削減に成功したことから，小型PCやモバイル端末への実装が可能となり，その結果，悲鳴検出システムの応用先が拡大したといえる．例えば，モバイル端末のアプリケーションとして提供されれば，各個人が所有する端末が通報装置となるため，高い犯罪抑止効果が得られる．その他の利用例として，防犯カメラと併用することで，悲鳴発生源に焦点を当て，より鮮明な証拠映像を捉えることも可能となる．