Demonstration of all-parametric vowel synthesis according to [1] and [2]
V1 – vowel /a/ by a female speaker using a constant LPC spectral envelope model, a constant sawtooth harmonic phase model, and a constant F0, in addition to a smooth fade-in and fade-out amplitude control:
V2 – adding F0 microvariations to V1:
V3 – replacing the sawtooth harmonic phase model by a constant harmonic phase model approximation to the original vowel sound on V2:
V4 – adding individual harmonic magnitude deviations relative to the constant LPC spectral envelope model (first 20 harmonics only) on V3:
V5 – adding synthetic noise floor according to an LPC spectral envelope model approximation to the original vowel sound (see [1]) on V4:
Original vowel sound (natural):
References:
[1] João Silva, Marco Oliveira, Aníbal Ferreira, “Flexible parametric implantation of voicing in whispered speech under scarce training data”, 28th European Signal Processing Conference, (EUSIPCO 2020), January 18-22, 2021, Amsterdam, The Netherlands
[2] Aníbal J. Ferreira, João Silva, Francisca Brito, Deepen Sinha, “Impact of a Shift-Invariant Harmonic Phase Model in Fully Parametric Harmonic Voice Representation and Time/Frequency Synthesis”,
International Conference on Acoustics, Speech, and Signal Processing (ICASSP2020), May 4-8, Barcelona, Spain, 2020