On finding the optimum excitation for LPC speech synthesis
Bishnu Saroop Atal, Nancy David · The Journal of the Acoustical Society of America · 1978
In synthesizing speech from LPC parameters, the excitation for voiced speech is generally approximated by a sequence of pulses spaced one pitch period apart. This idealization assumes that the different harmonics in the excitation have equal amplitudes and zero phases. The object of this study was to determine the best modification—particularly with respect to the phrase—of this simple pulse excitation for optimizing the quality of resultant synthetic speech. We systematically modified the amplitudes and phases of the harmonics in the excitation signal obtained by inverse filtering of the speech waveform. Our results show that replacing the actual harmonic amplitudes with constants—even when the phases are correct—produces audible distortion in the synthetic speech. Furthermore, exact reproduction of the amplitude spectrum alone is not sufficient for producing natural-sounding speech—the relative phases of different harmonics must also be considered. Preliminary results suggest that actual phases can be replaced by frequency-dependent but time-invariant phases without introducing significant distortion.