This was featured in ACM TechNews. There isn't much details, but I'm copying it here in case someone finds it interesting.

----------------------------------------------------------------------------------------------------------
Research to Improve Speech Recognition Software
Binghamton University (08/10/10) Coker, Rachel 

Binghamton University (BU) researchers are developing methods to improve the accuracy of automatic speech recognition technology. BU professor Stephen Zahorian is creating a multi-language, multi-speaker audio database that will be available for spoken-language processing research. "The challenge is to get speech recognition working better in real-life situations," Zahorian says. The database's speech samples come from publicly available sources such as YouTube. The BU team annotates each sample, creating a more detailed version of closed captioning, including time stamps and descriptions of background sounds. Once the transcription is complete, automatic speech recognition algorithms align the recording with the captions. The software then verifies and corrects errors in the time alignment. "We want the algorithms to extract just the most useful characteristics of the speech, not all of the possible data," Zahorian says. "That’s because more detail can actually hurt performance, past a certain point." He says the BU database will join others in the Linguistic Data Consortium, providing researchers with a new way to test theories with real-life speech samples.

http://discovere.binghamton.edu/faculty-spotlights/zahorian-3249.html

--------------------------------------------------------------------

--- (Edited on 8/13/2010 5:40 pm [GMT-0500] by RainCT) ---