15th International Congress of Phonetic Sciences (ICPhS-15)Barcelona, Spain |
Appropriate prosody modeling is crucial for natural sounding text to
speech synthesis, and accurate estimation of segmental durations greatly
contributes to this naturalness, not only by establishing phone durations,
but also because of the dependency of the intonation models on those
estimated durations.
This paper presents the modeling of segmental
durations performed for standard Basque using binary regression trees
and describes the experiments and results obtained when using different
predicting factors, as well as when testing several target variables
and phone groupings. Additionally, a subjective evaluation of the
duration model has been performed and is described together with the
final results.
Bibliographic reference. Navas, Eva / Hernáez, Inmaculada / Sánchez, Juan Maria (2003): "Predicting segmental durations for Basque using CARTs", In ICPhS-15, 2083-2086.