Speech enhancement using Fast Hartley Transform

In automotive hands free communication, speech is mostly corrupted by engine and other background noises. Engine noise can be removed using frequency domain adaptive filtering as proposed by Benesty [1, 2]. Subband speech processing can be used to further enhance the speech signal by suppressing residual and background noise, as proposed by Diethorn [3]. For both these algorithms, significant computational savings can be obtained by using faster time-to-frequency transformation techniques and reducing order of calculations in the frequency domain. Benesty and Diethorn [1, 3] have formulated the noise canceller and noise suppressor algorithms using Fast Fourier Transform (FFT). In this paper, we reformulate the problem using Fast Hartley Transform (FHT) for both the adaptive noise canceller and subband noise suppressor algorithms, which reduces the computational complexity to half that of FFT based implementation.