近年，遠隔会議やロボット聴覚等におけるハンズフリーrn音声入力を目的として，マイクロホンアレーを用いた音響rn信号処理技術が注目されている。マイクロホンアレーrnは，複数のマイクロホンを並べた収音系であり，マイクロホrnンの位置の違いを利用した空間信号処理が可能である。こrnの性質により，空間感度特性である指向特性において，特rn定の話者（目的音源）の方向に村する感度を一定に保ちながrnら，雑音源方向にヌルを形成することにより雑音環境下にrnおける目的音の高S/N受音が可能となる。すなわち，高rnS/N受音のためには指向特性が目的音源方向に鋭いピークrnをもつことが効果的である。%This paper presents a novel method for moving sound source localization and its performance evaluation in actual room environments. The method is based on the MUSIC (Multiple Signal Classification) which is one of the most high resolution localization methods. When using the MUSIC, a computation of eigenvectors of correlation matrix is required for the estimation. It needs often a high computational costs. Especially, in the situation of moving source, it becomes a crucial drawback because the estimation must be conducted at every the observation time. Moreover, since the correlation matrix varies its characteristics due to the spatial-temporal non-stationarity, the matrix have to be estimated using only a few observed samples. It makes the estimation accuracy degraded. In this paper, the PAST (Projection Approximation Subspace Tracking) is applied for sequentially estimating the eigenvectors spanning the subspace. In the PAST, the eigen-decomposition is not required, and therefore it is possible to reduce the computational costs. Several experimental results in the actual room environments are shown to present the superior performance of the proposed method.
展开▼
