14th International Congress of Phonetic Sciences (ICPhS-14)

San Francisco, CA, USA
August 1-7, 1999

Duration of Russian Consonants in Clusters: Data and Modeling for Concatenative Text-to-Speech System

Nina Zinovieva (1,2)

(1) Moscow State University, Moscow, Russia
(2) Lernout & Hauspie Speech Products, Boston, USA

Accurate modeling of consonant durations in TTS systems is very important for naturalness of synthetic speech, especially for languages with complex and frequent consonant clusters, like Russian. Consonant duration in multiple clusters impacts many perceptually significant features of speech, including such fundamental as rhythm. However, there are no systematic data on consonant duration in clusters. The goal of this research was to reveal the main tendencies of the consonant duration behavior in clusters with respect to the different factors, like stress, cluster position in a word/phrase, manner of articulation, consonant position in a cluster. The results of the study revealed that all these factors (and some others) are involved in modifying consonant duration in clusters, but their role depends on a particular combination of them. It's been shown also that different groups of consonants and, moreover, different parts of consonants have different levels of resistance to modifying factors.

Full Paper

Bibliographic reference.  Zinovieva, Nina (1999): "Duration of Russian consonants in clusters: data and modeling for concatenative text-to-speech system", In ICPhS-14, 2287-2290.