表題番号:2021C-737 日付:2022/04/04
研究課題音環境の認識と理解のための革新的マイクロホンアレー基盤技術の研究
研究者所属(当時) 資格 氏名
(代表者) 理工学術院 大学院情報生産システム研究科 特任教授 牧野 昭二
研究成果概要

This research explores whether the newly proposed online algorithm that jointly optimizes weighted prediction error (WPE) and independent vector analysis (IVA) works well in separating moving sound sources in reverberant indoor environments. The moving source is first fixed and then rotated 60 degrees in a room at a speed of less than 10 cm/s, while the other remains fixed. Through the comparison of the online-AuxIVA, online-WPE+IVA (separate), and online-WPE+IVA (joint) algorithms, we can conclude that the online-WPE+IVA (joint) method has the best separation performance when the sources are fixed, but online-WPE+IVA (separate) is more stable and has better performance when removing moving sources from the mixed sound.