The paper presents the decomposition of the speech signal into two modulated components, namely the envelope and the instantaneous frequency (IF). For this purpose, a bandpass signal is first represented as a sum of sinusoidal signals and then, by using the roots of the polynomial with complex coefficients of the signal, it is transformed in a product of elementary signals. This representation allows an easier computation of the envelope and of the IF. In order to eliminate the computation of the roots, a LPSD (linear prediction in spectral domain) algorithm is used. Thus, the two components of the signal can be computed. First, the method is applied to a synthesized signal, and then to a speech signal.
Cite as: Mischie, S. (2004) Using linear prediction in spectral domain to decomposition speech into modulated components. Proc. 9th Conference on Speech and Computer (SPECOM 2004), 203-209
@inproceedings{mischie04_specom,
author={Septimiu Mischie},
title={{Using linear prediction in spectral domain to decomposition speech into modulated components}},
year=2004,
booktitle={Proc. 9th Conference on Speech and Computer (SPECOM 2004)},
pages={203--209}
}