A Preprocessing for Sound Source Separation Using Complex Weighted Sum Circuits
A Preprocessing for Sound Source Separation Using Complex Weighted Sum Circuits
复制标题
DOI：
10.1109/ispacs48206.2019.8986346
复制
发表时间：
2019-12
期刊：
影响因子：
--
通讯作者：
Shun Nishimaki;K. Suyama
中科院分区：
文献类型：
--
作者：
Shun Nishimaki;K. Suyama
本文研究了一种形成指向性的声源分离预处理方法。采用复加权和电路形成指向性。此外，采用多重复加权和电路扩展抑制区间。但是，即使已经扩大了抑制间隔，由于混响，干扰声可能仍然存在。在该方法中，作为分离的预处理，对只包含干扰声的帧进行选择性抑制。为了检测只有干扰声的帧，使用源的DOA（到达方向）。通过这种预处理，可以抑制形成指向性不能抑制的干扰声，提高分离性能。
In this paper, a preprocessing method for sound source separation by forming the directivity is studied. A complex weighted sum circuit is used to form the directivity. Furthermore, multiple complex weighted sum circuits are used to spread the suppression interval. However, even if the suppression interval has been spread, the interference sound may be remained due to the reverberation. In the proposed method, as a preprocessing for separation, the frames including just the interference sound is suppressed selectively. To detect the frames having just only the interference sound, the DOA (Direction-Of-Arrival) of the source is used. Applying this preprocessing, the interference sound that can not be suppressed by forming the directivity can be suppressed, and an improvement of the separation performance is expected.