Sixth ISCA Workshop on Speech Synthesis

Bonn, Germany
August 22-24, 2007

Voice Building from Insufficient Data - Classroom Experiences with Web-Based Language Development Tools

John Kominek, Tanja Schultz, Alan W. Black

Language Technologies Institute, Carnegie Mellon University, Pittsburgh, USA

To make the goal of building voices in new languages easier and more accessible to non-experts, the combined tasks of phoneme set definition, text selection, prompt recording, lexicon building, and voice creation in Festival are now integrated behind a web-based development environment. This environment has been exercised in a semester-long laboratory course taught at Carnegie Mellon University. Here we report on the students' efforts in building voices for the languages of Bulgarian, English, German, Hindi, Konkani, Mandarin, and Vietnamese. In some cases intelligible synthesizers were built from as little as ten minutes of recorded speech.

Full Paper     Poster (pdf)

Sound Examples  
German:
01   02   03   04   05   06   07   08   09   10   11   12   13   14   15   16   17   18   19   20   21   22   23   24   all  
Hindi:
01   02   03   04   05   06   07   08   09   10   11   12   13   14   15   16   17   18   19   20   21   22   23   24   all  

These files contain the German and Hindi test wavefiles described in Section 5.1 of the paper.

Bibliographic reference.  Kominek, John / Schultz, Tanja / Black, Alan W. (2007): "Voice building from insufficient data - classroom experiences with web-based language development tools", In SSW6-2007, 322-327.