A nonlinear adaptive noise canceller with multiple reference channels for speech enhancement using both bone-and air-conducted measurements
Yegui Xiao, Tao Bai, Ran Xiao, Yaping Mat · 2018
When the background noise is large compared to the air-conducted speech signal, performance of most existing speech denoising techniques will significantly degrade. In recent years, speech recovery techniques using corrupted air-conducted speech and relatively clean bone-conducted signals have been proposed that utilize an adaptive noise canceller (ANC) framework. In this paper, we propose a multi-channel nonlinear ANC to perform the denoising task, that consists of a single primary and multiple reference channels. The ANC is nonlinear and contains an FIR filter and a generalized functional link artificial neural network. Application to real air-and bone-conducted measurements is carried out to demonstrate the performance and advantages of the proposed system in recovering the high-frequency components of the original speech signal.