作者: A.H. Nour-Eldin , T.Z. Shabestary , P. Kabal
DOI: 10.1109/ICASSP.2006.1660588
关键词:
摘要: In this paper, we investigate the effect of temporal correlation on dependence between speech narrow and high frequency bands covering 0.3–3.4 kHz 3.7–8 ranges, respectively. We follow technique using Gaussian mixture modelling spectral envelopes represented by Mel-frequency cepstral coefficients. The disjoint is quantified through mutual information (MI) its ratio to highband entropy. Speech exhibits considerable that not explicitly accounted for static parametrization envelopes. Including memory in (through delta features) incorporates such modelling, hence, MI gains are be expected resulting bandwidth extension with better performance. Results show exploiting features can increase certainty about (ratio entropy) as much 216% relatively, corresponding an absolute 12%.