Title: Recent improvements on ARTIC: czech text-to-speech system
Other Titles: NejnovÄ›jší vylepšenĂ­ ÄŤeskĂ©ho TTS systĂ©mu ARTIC
Authors: Matoušek, Jindřich
Romportl, Jan
Tihelka, Daniel
Tychtl, Zbyněk
Citation: MATOUĹ EK, JindĹ™ich; ROMPORTL, Jan; TIHELKA, Daniel; TYCHTL, ZbynÄ›k. Recent improvements on ARTIC: czech text-to-speech system. In: Proceedings of ICSPL 2004: 8th Annual Conference of the International Speech Communication Association 2004, Jeju Island, Korea, 4-8 October 2004. [Baixas]: ISCA, 2004, p. 1933-1936.
Issue Date: 2004
Publisher: ISCA
Document type: článek
article
URI: http://www.kky.zcu.cz/cs/publications/MatousekJ_2004_RecentImprovements_1
http://hdl.handle.net/11025/17027
Keywords: syntĂ©za Ĺ™eÄŤi;automatickĂ© vytvářenĂ­ inventáře akustickĂ˝ch jednotek;modelovánĂ­ rázu;shlukovánĂ­ Ĺ™eÄŤovĂ˝ch jednotek;automatická segmentace Ĺ™eÄŤi;datovÄ› řízenĂ© generovánĂ­ prozodie
Keywords in different language: speech synthesis;automatic acoustic unit inventory creation;glottal stop modelling;speech unit clustering;automatic speech segmentation;data-driven prosody generation
Abstract: ÄŚlánek prezentuje nejnovÄ›jší vylepšenĂ­ systĂ©mu ARTIC, modernĂ­ho ÄŤeskĂ©ho korpusovÄ› orientovanĂ©ho TTS systĂ©mu. ProtoĹľe jsme pouĹľili statistickĂ˝ přístup (skrytĂ© Markovovy modely) k vytvoĹ™enĂ­ inventáře akustickĂ˝ch jednotek, vylepšenĂ­ se tĂ˝kala všech jeho komponent. VylepšenĂ˝m modelovánĂ­m, shlukovánĂ­m, a segmentacĂ­ akustickĂ˝ch jednotek jsme dosáhli zvýšenĂ© srozumitelnosti vĂ˝slednĂ© Ĺ™eÄŤi. Navrhli jsme rovněž 2 přístupy ke generovánĂ­ prozodickĂ˝ch charakteristik a zĂ­skali tak vyšší pĹ™irozenost syntetickĂ© Ĺ™eÄŤi. Abychom zvýšili i plynulost vytvářenĂ© Ĺ™eÄŤi, navrhli jsme rovněž schĂ©ma využívajĂ­cĂ­ vĂ­ce realizacĂ­ kaĹľdĂ© Ĺ™eÄŤovĂ© jednotky s vĂ˝bÄ›rem nejvhodnÄ›jšího kandidáta on-line. ZmĂ­nĂ­me takĂ© alternativnĂ­ metodu vytvářenĂ­ Ĺ™eÄŤi využívajĂ­cĂ­ harmonickĂ˝ model a model šumu. ImplementacĂ­ nÄ›meckĂ©ho a slovenskĂ©ho jazykovĂ©ho modulu (vedle 2 ÄŤeskĂ˝ch hlasĹŻ) jsme navĂ­c vytvoĹ™ili dĹŻleĹľitĂ˝ krok smÄ›rem k vĂ­cejazyÄŤnosti našeho TTS systĂ©mu ARTIC.
Abstract in different language: This paper presents recent improvements on ARTIC - the modern Czech corpus-based text-to-speech system. As a statistical approach (using hidden Markov models) was applied to create an acoustic unit inventory, several improvements concerning acoustic unit modelling, clustering and segmentation have been accomplished to increase the intelligibility of the resulting speech. Two approaches to the generation of prosodic features were also proposed and implemented to increase the naturalness of synthetic speech. To produce as smooth synthetic speech as possible, a multiple unit instance scheme with on-line unit candidate selection was proposed as well. Our work on an alternative harmonic/noise-based speech production method is also mentioned. In addition, an important step towards multilinguality was achieved as German and Slovak language modules were implemented besides two Czech voices within the framework of ARTIC TTS system.
Rights: Â© JindĹ™ich Matoušek - Jan Romportl - Daniel Tihelka - ZbynÄ›k Tychtl
Appears in Collections:Články / Articles (MMI)
Články / Articles (KKY)
Články / Articles (NTIS)
Články / Articles (KIV)

Files in This Item:
File Description SizeFormat 
MatousekJ_2004_RecentImprovements_1.pdfPlný text553,47 kBAdobe PDFView/Open


Please use this identifier to cite or link to this item: http://hdl.handle.net/11025/17027

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.