Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

ARTIC: A New Czech Text-To-Speech System Using Statistical Approach to Speech Segment Database Construction

Jindrich Matousek, Josef Psutka

University of West Bohemia, Department of Cybernetics, Plzen, Czech Republic

This paper presents ARTIC1, a brand-new Czech text-to-speech (TTS) system. ARTIC (ARtificial Talker In Czech) is a concatenation-based system that consists of three main, relatively independent, components: speech segment database, text analyzer and speech synthesizer. A statistical approach to speech segment database construction is used: Hidden Markov models are employed to model triphones on the basis of the large speech corpus and to segment the corpus into triphonebased speech units - basic speech units used by the synthesizer. A speech segment selection algorithm is described to choose the representative instance of each speech unit from the segmented speech corpus. A text processing module converts the written text at the input of TTS system to the sequence of phones - basic phonetic units needed to describe the pronunciation of the input text - and prosodic marks. Finally, speech processing is performed using two versions of a PSOLA algorithm.

Full Paper

Acoustic Example #1    Acoustic Example #2    Acoustic Example #3    Acoustic Example #4

Bibliographic reference.  Matousek, Jindrich / Psutka, Josef (2000): "ARTIC: a new Czech text-to-speech system using statistical approach to speech segment database construction", In ICSLP-2000, vol.4, 612-615.