Korean TTS (Text-to-speech) project

Part of an SBC/Ameritech -funded project developing Text-to-speech synthesis systems in Festival for Cantonese, Puerto Rican Spanish, and Seoul Korean.

Started in Autumn 2001.

Funding ended in Spring 2002.

However, research goes on..., as always.

Professors in charge: Mary Beckman and Chris Brew .

Research assistants: Peggy Wong (Cantonese), Michelle F. Ramos-Pellicia (Puerto Rican Spanish), and Kyuchul Yoon (Seoul Korean).

New members joined the Korean part of the project in September 2002:  Eunjong Kong and Hyunsook Shin.

Festival Speech Synthesis System (Overview , Downloads, Mailing list , FAQ ):

• Documentation: The Festival Speech Synthesis Systemn; System documentation . (postscript version) .  Edition 1.4, for Festival version 1.4.2.  25th July 2001 by Alan W. Black, Paul Taylor and Richard Caley at Speech Group at Carnegie Mellon University .
• Documentation: Building Voices in the Festival Speech Synthesis System; Processes and issues in building speech synthesis voices.   Edition 1.2:beta, for Festival version 1.4.1.  10th July 2000 by Alan Black and Kevin A. Lenzo.  (postscript version)
• Documentation: Edinburgh Speech Tools Library: System documentation .  Edition 1.2 for version 1.2.0.  15th June 1999 by Paul Taylor, Richard Caley, Alan W. Black and Simon King at Centre for Speech Technology , University of Edinburgh .
• Documentation: Limited Domain Synthesis by Alan Black and Kevin A. Lenzo. Examples. Example databases.
• Festival Speech Architecture .
• EMU Speech Database System . Downloads .
• Documentation: The EMU Speech Database System .  Version 1.2 by Steve Cassidy at Speech Hearing and Language Research Centre , Macquarie University .
• Links to related projects .
• On-line publications for Festival and speech synthesis topics .
• CSTR recent publications .
• Flite: a small fast run-time synthesis engine by Alan W. Black and Kevin A. Lenzo, published at The 4th ISCA Workshop on Speech Synthesis .

Project progress for Korean:

• Letter-To-Sound rules for Korean. 2002. Paper presented at IEEE2002 Workshop on speech synthesis. Santa Monica, California.
• LTS augmentation: Model trained from labeled Sino-Korean words.
• Sino-Korean words: tensification, n-l/l-n words, s-borrowings.
• Korean accentual phrase boundary prediction: Model trained from K-ToBI labeled parsed read corpus.
  
• Syntactic parsing of selected articles from Korean Newswire corpus. (in collaboration with UPenn)
• K-ToBI labeling of read Newswire corpus.
• Features that will be used in training: parts-of-speech, phrases, and punctuations.
• Relevant materials can be found here.
• Duration modeling from the same corpus that was used in AP boundary prediction.
  
• Prosodically-conditioned diphone synthesis & Eojeols: ( diphone eojeols, tentative )
• IP initial vs. AP initial vs. PW medial