Retraining and updating speech models for speech recognition

摘要: A technique is provided for updating speech models recognition by identifying, from a class of users, data predetermined set utterances that differ stored at least amount. The identified similar the users collected and used to correct models. As result, corrected are closer match than were subsequently updated with provide improved users. For example, may be processed central database returned, via suitable communications channel (e.g. Internet) individual user sites update apparatus those sites.