Multiple beamforming with source localization based on CSP analysis

摘要

In this paper, a method is investigated for using an array of microphones to capture a high-quality recording of voice under reverberation. In the past, multiple beamforming has been proposed by Flanagan's group as a method for efficiently suppressing reverberation. However, because this multiple beamforming method computes the arrival positions of the sound and its reflections based on the prior knowledge of the shape of the room and the position of the target sound source, it is not suited to practical applications. In this paper, a method of multiple beamforming is proposed in which the source positions of the sound and its reflections are localized without using prior knowledge of the shape of the room or the position of the target sound source. The effectiveness of the proposed method was evaluated in a simulation experiment using the image method and in an experiment in an actual hallway environment. First, an experiment was run to localize the positions of the sources of the sound and its reflections under a condition in which both the shape of the room and the source position of the sound were unknown. As the result, it was found that the localization error for the source position of the sound was small. Furthermore, although the localization errors for the source positions of the reflections were large, the localized reflection paths were found to be approximately equal to the true reflection paths. Next, the output signal from the proposed method was evaluated based on SNR. The result demonstrated that multiple beamforming is more effective than single beamforming.