CODIFICAÇÃO DE SINAIS DE VOZ HUMANA POR DECOMPOSIÇÃO EM COMPONENTES MODULANTES

Abstract

This work proposes an speech signal encoder variation based on two concepts: the formants and the modulating components of the speech signal. The method suggested for the codification extracts the modulating components (instantaneous amplitude and frequency) to be transmitted. The method is based on the fact that the transmission of the speech can be substituted by the transmission of its AM-FM modulating components (amplitude modulation - frequency modulation). Thus, to send such components, the LPC (linear predictive coding) method is used to determine the frequencies that correspond to the first four formants of the speech spectrum within a 4 kHz band. Then, through a modified Gabor s wavelet function, four narrow bands are filtered around the formants. Finally, the properties of the Hilbert transform are used to determine the modulating components of the filtered bands, in other words, the instantaneous amplitudes and frequencies. The final result is the codification of eight signals in which four of them correspond to the instantaneous amplitudes and the other four correspond to the instantaneous frequencies. It is also presented a recovery of human speech where tests of intelligibility of the samples are applied after their respective recoveries. The results obtained showed that the method is a promising technique to be implemented in actual applications.