Research

Jargon: MT output may have points of disfluency, but synthesizers usually don't deal with them explicitly. That can make the synthesized speech unintelligible. I am trying to see how synthesis can be made more understandable when the input is disfluent MT text. This work fits the context of lecture translation in the PT-Star project.

Basically: Imagine a professor delivering a lecture. A computer translates it live to you, in your language. The computer makes mistakes and can be difficult to understand. I am trying to make it better.

My Advisor: Alan W Black

Education

MS (Language Technologies), Carnegie Mellon University
2007--2009.       QPA: 3.86

B. Tech (Information Technology), Indian Institute of Information Technology, Allahabad
2002--2006.       GPA: 9.83/10

Voices

English:

My Festvox voice for English, built on the Europarl corpus will be available for download soon. There is an online DEMO.

Marathi:

I am building a voice that will synthesize Marathi text. It will take a while. The current estimated progress is 1%.

Publications