|   CMU Speech Software   |   CMU Speech Group   |  

FestVox Download
Festival Download
Voice Demos
Limited Domain

Example Databases
    KED timit
    KAL diphone
    RAB diphone
    Time ldom
    Weather ldom
    Communicator ldom

Mailing Lists
Search Documents
Contributed parts

Speech Synthesis Databases
In order to make building voices easier we offer speech synthesis databases which serve as examples to the techniques described in the festvox document.

General Databases

  • CMU ARCTIC, 18 single speaker speech databases with around 1200 phonetically balanced uttrances.
  • CMU INDIC, 13 single speaker speech databases, Bengali (1), Gujarati (3), Hindi (1), Kannada (1), Marathi (2), Panjabi (1), Tamil (1), and Telugu (3), often with English recordings too.
  • CMU FAF, 107 paragraphs (15,000 words) of single speaker monologues with interesting prosody. Based on Aesop's fables and country descriptions in the CIA world fact book.
  • CMU SIN, speech in noise: speech recorded while noise is playing in the speakers ear's (and when not).
  • CSTR US KED timit University of Edinburgh's male US TIMIT, 452 phonetically balanced utterances.

Limited Domain Databases

Diphone Databases

MBROLA voices and binaries (US mirror)

  • A US mirror of The MBROLA projects wide range of pre-built diphone databases for many languages and binaries for the mbrola program itself for many platforms.
CMU/LTI This page is maintained by Alan W Black (awb@cs.cmu.edu)
Festvox is a project within LTI at Carnegie Mellon University