Convolutional Bias Removal Based on Normalizing the Filterbank Spectral Magnitude

Abstract

In this letter, a novel convolutional bias removal technique is proposed. The proposed method is based on scaling the filterbank magnitude by the average of filterbank magnitude over time. The relation between the cepstral mean normalization (CMN) and proposed algorithm is derived. The experimental results show that the proposed algorithm is more robust than the CMN for both convolutional bias and additive noise. For example, the proposed method reduced the equal error rate by 5.66% and 10.16% on average for the convolutional bias and 12-dB additive noise, respectively.