能動音響制御と音声分離手法を用いた音声明瞭化

抄録

Active noise control (ANC) is a technique used to reduce environmental noise in noisy places such as office or aircraft cabin. ANC reduces the overall input signal even when it has necessary information such as conversation as well as noise. In other words, conventional ANC has the disadvantage of reducing the required sound. On the other hand, there are scenes where conversation is desired even in a noisy environment. However, since the entire input signal is reduced, it cannot be solved even by using conventional ANC. We propose speech separation ANC system which separate announce and noise then control only noise component by ANC. To separate noise from the input signal, we focus on the signal separation method of spectral subtraction method and harmonic extraction method. These methods are simple and require a small computational load but have the disadvantage of leaving an estimated noise component. We also propose two signal separation methods which remain less noise component. In this paper, we verify the effectiveness of the proposed signal separation method while comparing it with the conventional signal separation method. Furthermore, the sound attenuation performance of the proposed method is confirmed by a control experiment in an anechoic room.