A Deep Learning Method to Multi-Channel Active Noise Control
(3 minutes introduction)
| Hao Zhang (Ohio State University, USA), DeLiang Wang (Ohio State University, USA) |
|---|
This paper addresses multi-channel active noise control (MCANC) on the basis of deep ANC, which performs active noise control by employing deep learning to encode the optimal control parameters corresponding to different noises and environments. The proposed method trains a convolutional recurrent network (CRN) to estimate the real and imaginary spectrograms of all the canceling signals simultaneously from the reference signals so that the corresponding anti-noises cancel or attenuate the primary noises in an MCANC system. We evaluate the proposed method under multiple MCANC setups and investigate the impact of the number of canceling loudspeakers and error microphones on the overall performance. Experimental results show that deep ANC is effective for MCANC in various scenarios. Moreover, the proposed method is robust against untrained noises and works well in the presence of loudspeaker nonlinearity.
