Patent Review

ADAPTATION OF A SPEECH RECOGNITION SYSTEM ACROSS MULTIPLE REMOTE SESSIONS WITH A SPEAKER

Patent No.: US 6,766,295 B1

Inventors: Hy Murveit, Portola Valley, CA (U S); Ashvin Kannan, Redwood City, CA (Us)

Publication Date:Jul. 20, 2004

Filing Date : May 10’ 1999

A speaker can remotely access a speech recognition system. The medium of communication can be a telephone. An acoustic model is used to initialize an independent speaker in a remote session for the first time.  Upon the termination of the session, the speech data of the speaker is stored in memory with an independent identification key . During a subsequent remote session, the speaker is identified on the basis of his speech and the corresponding acoustic model is altered. New samples are taken each time the user speaks. The stored acoustic model is modified each time and the result made more accurate.

Even if the speaker uses the remote session for a relatively small amount of time, modifications are made in the acoustic model and the accuracy of recognition is improved. Thus different users can use multiple remote sessions and their speech data can be stored and modified separately.