Click Here for the full article.

Use the open source Sphinx-4 speech-recognition package to capture letters and numbers from spoken conversations in near real time to create notes. Employ a custom Sphinx-4 dictionary file to extract likely matches to spoken letters and numbers. Process the text for higher order values such as phone numbers and acronyms, and create a meeting annotator through search-engine lookups and local databases.

Original link at IBM developerworks. Please use the cached copy if the link is dead.
Date