Festival and Speech Synthesis at CMU

This page describes current projects in speech synthesis in the speech group and the Language Technologies Institute at Carnegie Mellon University.

Most of this work is within the framework of Edinburgh University's Festival Speech Synthesis System.

Goal

• Making speech synthesis as natural, flexible and efficient as human speech

Synthesis Research Areas

• The Festvox project: automating the processes involved in building synthetic voices for new languages
  Documentation, tools, scripts and example databases are available at http://www.festvox.org. This release festvox-1.1-beta.tar.gz includes:
  • Support for designing, recording and autolabelling diphone databases
  • Support for designing, recording and autolabelling unit selection databases
  • Building simple limited domain synthesis engines
  • Support for building rule driven and data driven prosody models
  • Lexicon and building letter to sound rule support
  • Predefined scripts for building new US (and UK) English voices
• Unit selection techniques for "limited domain" synthesis
  See http://www.festvox.org/ldom/index.html for discussion and examples.
• Prosodic models trained from dialog speech and other speaker characteristic and style
• light-weight real-time synthesis
  • Small, fast portable synthesis engine based on Festival technology using compilation methods to give efficent scalable voices. See CMU Flite.

Synthesis Resources

• Speech synthesis demos of Festival and CMU related synthesis projects.
• Festvox project documentation, tools scripts and example speech databases for building new synthetic voices
• Synthesis databases speech databases for using synthesis research, diphones, timit and domain dependent.
• University of Edinburgh's Festival Speech Synthesis System (at Edinburgh)
• Festival 1.4.1 Online Manual (at CMU)
• Festival distribution (at CMU)

For more details please contact Alan W Black of LTI at CMU.