On Smoothing and Enhancing Dynamics of Pitch Contours Represented by Discrete Orthogonal Polynomials for Prosody Generation

Chen-Yu Chiang


This paper presents a new pitch contour generation algorithm for statistical syllable-based logF0 generation models which represent logF0 contours of syllables by coefficients of discrete orthogonal polynomials, i.e. orthogonal expansion coefficients (OECs). The conventional statistical logF0 models can generate smooth pitch contour within a syllable because of the continuity property of polynomials. However, the models do not ensure to produce continuous and smooth logF0 contours in the proximity of syllable junctures. Besides, dynamic range of the generated logF0 contours is generally smaller than the one of real speech. The above two shortcomings would result in unnatural and monotonous prosody. To overcome these shortcomings, juncture-smooth and dynamics-enhancing OEC generation algorithms are hence proposed in this paper. Analysis on the generated logF0 contours by the proposed algorithm shows some improvements in logF0 smoothness at syllable junctures and enhanced logF0 dynamic range. In addition, a perceptual evaluation of the logF0 contour generated by the proposed algorithm shows an improvement in naturalness of the synthesized speech.


DOI: 10.21437/Interspeech.2016-409

Cite as

Chiang, C. (2016) On Smoothing and Enhancing Dynamics of Pitch Contours Represented by Discrete Orthogonal Polynomials for Prosody Generation. Proc. Interspeech 2016, 2303-2307.

Bibtex
@inproceedings{Chiang2016,
author={Chen-Yu Chiang},
title={On Smoothing and Enhancing Dynamics of Pitch Contours Represented by Discrete Orthogonal Polynomials for Prosody Generation},
year=2016,
booktitle={Interspeech 2016},
doi={10.21437/Interspeech.2016-409},
url={http://dx.doi.org/10.21437/Interspeech.2016-409},
pages={2303--2307}
}