Modélisation 2D (« fréquence-temps ») des amplitudes spectrales
Mohammad Firouzmand, Laurent Girin.
This paper presents a method for modeling the spectral amplitude parameters of speech signals in “two dimensions” (2D). It consists in two cascaded modeling: the first one along the frequency axis is usual, since it consists in modeling the log-scaled spectral envelope with a sum of Discrete Cosine (DC) functions. The second one, along the time axis, consists in modeling the trajectory of the envelope DC parameters by another similar DC model. An iterative algorithm that optimally fits this 2D-model, taking into account perceptual criterions, is proposed. This approach is shown to provide an efficient representation of speech spectral amplitude parameters in terms of coefficient rates, while providing good signal quality, opening new perspectives in very-low bit-rate speech coding.