Sixth European Conference on Speech Communication and Technology
The paper presents a more accurate model of speech synthesis which can be applied to the low bit-rate vocoders producing synthetic speech of high quality. Speech is described as a combination of the periodic and the noise components within given time frame. The periodic spectrum that appears in the voiced bands is a set of pitch harmonics, the amplitude and the frequency of which linearly changes from one window of analysis to another. The noise component is described as a difference between the input speech signal and the synthesized periodic spectrum. The system of functions employed in the analysis procedure is determined taking into account the changes in frequencies of pitch harmonics. This allows to reproduce the periodic spectrum with minimal distortion and separate it from the noise component. The paper includes results of experiments showing advantages of the suggested model in comparison with the existing ones.
Full Paper (PDF) Gnu-Zipped Postscript
Bibliographic reference. Sercov, Valentin V. / Petrovsky, Alexander A. (1999): "An improved speech model with allowance for time-varying pitch harmonic amplitudes and frequencies in low bit-rate MBE coders", In EUROSPEECH'99, 1479-1482.