Third International Workshop on Models and Analysis of Vocal Emissions for Biomedical Applications (MAVEBA 2003)
In this paper, we are interested in multichannel speech denoising in the context of mobile communications. The conventional method exploits the similarity between the avalaible observations in the sense of the coherence function, measured in the Fourier domain. In this work, we alleviate the limitations of this approach by assessing the coherence function in the Modulated Lapped Transform (MLT) domain. Indeed, the MLT allows to take into account the local statistics of the underlying speech signal. Experimental simulations indicate the outperformance of the proposed method w.r.t. the conventional method: some distortions are reduced and the intelligibility is enhanced.
Full Paper (reprinted with permission from Firenze University Press)
Bibliographic reference. Helaoui, L. / Jebara, S. Ben / Benazza-Benyahia, A. (2003): "A two-channel speech denoising method combining wavepackets and frequency coherence", In MAVEBA-2003, 57-60.