独立成分分析を用いた車内音寄与分離手法の基礎検討

-複数音源を持つロードノイズ，ウィンドノイズ成分音の分離精度向上法-

抄録

In this study, we considered a contribution separation technique that separates road and wind noise contributions to vehicle interior noise without using any sound source signals in case each noise was generated by multiple sound sources. In the experiment, road, wind or mixed noise sources were reproduced from four speakers positioned at different points. The mixed interior noises were recorded at two separate positions surrounded by the speakers. A contribution separation technique using independent component analysis (ICA) was then applied for obtaining the contribution at a position (driver seat position). Firstly, ICA was applied to the recorded interior noise at the driver seat (target noise) and the noise at the assistant seat position (reference noise). The result showed that each contribution could not be correctly obtained due to the low accuracy at the low frequency and the permutation problem, in which the relationship between the calculated and actual contributions was not retained with changes in frequency. Next, we changed the reference signal to the other noise having low correlation with the target noise to improve the accuracy and formulated the permutation solution using the correlation of calculated contributions of dominant independent component. By applying this technique, the ICA method with the permutation solution procedure correctly calculated the road and wind noise contributions.