Tuesday, June 29, 2010

Random thoughts to improve voice recognition

As stated once more, and this time on this ITWorld article, relative to Google making Google Voice available to SMB, improvement still has to be done on the voice recognition topic.

More and more editors already did add a voicemail to email transcript service. A service, of which precision is not really the main concern: sometimes even numbers are not correctly recognized!

Just a thought on that topic, maybe researchers could:
- add several noise filters
- recognize the language
- recognize the accent
- build a recognition model for each language and for each accent, or at least a global model for each language, and a global one for each accent
- train these models on a representative set

No comments:

Post a Comment