A high performance speech enhancement algorithm based on double-channel adaptive noise cancelling

Abstract

By employing the noise correlations between the main channel and the reference channel, a new adaptive filtering algorithm based on double-channel adaptive noise canceling (ANC) is proposed for speech enhancement. The proof of the new algorithm is also presented, and it is shown that its optimal weight vector is the solution of generalized Wiener equation. Its application is then discussed under additive white noise environment with low signal-to-noise ratio (SNR) ranging from 0-10 dB. Experimental results are as follows: compared with NLMS algorithm, the proposed algorithm can improve the SNR and PESQ score by 14 dB and 1.13, respectively, when the input noise containing crosstalk is assumed in the reference channel. Without the assumption, the SNR and PESQ score improvement become 5 dB and 0.83, respectively.