Improvement of virtual sound localization by reinforcement learning and EEG

2014 Fiscal Year Final Research Report

• Project/Area Number: 24650104
• Research Category: Grant-in-Aid for Challenging Exploratory Research
• Allocation Type: Multi-year Fund
• Research Field: Sensitivity informatics/Soft computing
• Research Institution: Nagaoka University of Technology
• Principal Investigator: WADA Yasuhiro 長岡技術科学大学, 工学(系)研究科(研究院), 教授 (70293248)
• Co-Investigator(Kenkyū-buntansha): OHISHI Kiyoshi 長岡技術科学大学, 工学部, 教授 (40185187) ; KUDO Akihiro 苫小牧工業高等専門学校, 工学部, 准教授 (80455097) ; NAMBU Isao 長岡技術科学大学, 工学部, 助教 (40553235)
• Project Period (FY): 2012-04-01 – 2015-03-31
• Keywords: 仮想音源 / 頭外音像定位 / 定位精度 / 強化学習

Outline of Final Research Achievements

Virtual sound localization is a technology that allows the user to perceive external sound sources using earphones. The exact sound image localization require an head-related transfer function (HRTF) for each listener.
To solve these problems, in this study, without HRTF measurements, and are intended to acquire the equivalent of the measured individual HRTF. To do so, HRTF is approximated using the Autoregressive Moving Average (ARMA) model, and for the coefficient was using a method of learning by unsupervised reinforcement learning. We used dummy head imitating the upper body of the person at an initial value of the learning, then, dynamically varied from a common HRTF, aimed to acquire each individual HRTF. We were verified whether it is possible to generate the HRTF for the purpose by the simulation. Thereafter, the listener experiment was verified the HRTF localization accuracy of which acquired by the simulation.

Free Research Field

計算論的神経科学