Bibliografické údaje¶
Název | Pitch Contours as Predictors of Audible Concatenation Artifacts |
Autor | Legát, M., Matoušek, J. |
Typ publikace | Článek ve sborníku konference (Conference Proceedings Citation Index, Thomson Reuters) |
Sborník | Lecture Notes in Engineering and Computer Science: Proceedings of The World Congress on Engineering and Computer Science (WCECS 2011) |
Místo konání | San Francisco, USA |
Strana | 525-529 |
Rok | 2011 |
ISBN | 978-988-18210-9-6 |
Abstrakt¶
This paper deals with the traditional problem of the occurrence of audible discontinuities at concatenation points at diphone boundaries in the concatenative speech synthesis. While most of the related studies put stress on the spectral component, we focused on the pitch contours and their role as predictors of the discontinuities. To measure the amount of information contained in the pitch contours, we trained SVM classifiers using perceptual data collected in listening tests. The results have shown that the fine grained pitch contours extracted from a vicinity of the concatenation points carry enough information for classifying continuous and discontinuous joins with a high accuracy.