Building speech models for new words in a multi-word utterance

United States Patent m

Roberts

[54] BUILDING SPEECH MODELS FOR NEW WORDS IN A MULTI-WORD UTTERANCE

[75] Inventor: Jed M. Roberts, Newton, Mass.

[73] Assignee: Dragon Systems, Inc., Newton, Mass.

[21] Appl. No.: 548,409
[22] Filed: Oct. 26,1995

[51] IntCl.6 G10L5/06

[52] U.S. CI 704/254; 704/243; 704/253;

704/270

[58] Field of Search 395/2.62.2.63,

395/2.64; 704/243, 253, 254, 270

[56] References Cited

U.S. PATENT DOCUMENTS

4,665,548 5/1987 Kahn 704/237

4,688,195 8/1987 Thompson et al 395/12

4,776,016 10/1988 Hansen 704/275

4,783,803 11/1988 Baker etal 704/252

4,803,729 2/1989 Baker 704/241

4,805,218 2/1989 Bamburg et al 704/241

4,805219 2/1989 Baker 704/241

4,827£20 5/1989 Zeinstia 704/275

4,829,576 5/1989 Porter 704/235

4,837,831 6/1989 Giffick et al 704/245

4,866,778 9/1989 Baker 704/254

4,903,305 2/1990 Gillick et al 704/245

4,914,703 4/1990 Gillick 704/245

4,931,950 6/1990 Isle et al 395/11

4,962,535 10/1990 Kimaraetal 704/252

4,984,177 1/1991 Rondel et al 704/277

5,027,406 6/1991 Roberts et al 704/244

[ii] Patent Number: 5,765,132 [45] Date of Patent: Jun. 9, 1998

5,036,538 7/1991 Oken et al 704/233

5,086,472 2/1992 Yoshida 704/255

5,095,508 3/1992 Fujimoto 704/247

5,127,055 6/1992 Larkey 704/244

5,170,432 12/1992 Hackbarth et al 704/254

5,202,952 4/1993 Gillick et al 704/200

5,231,670 7/1993 Goldhoretal 704/275

OTHER PUBLICATIONS

U.S. Patent Application 08/382,752, "Apparatuses and Methods for Training and Operating Speech Recognition Systems" Joel M. Gould et al., filed Feb. 1. 1995. U.S. Patent Application, "Speech Recognition." Serial No. 08/521.543, Gregory J. Gadbois, filed Aug. 30, 1995. U.S. Patent Application, "Speech Recognition," Serial No. 08/559,190, Gregory J. Gadbois, filed Nov. 13, 1995. U.S. Patent Application "Continuous Speech Recognition of Text and Commands," Serial No. 08/559,207, Joel M. Gould et al., filed Nov. 13, 1995.

Dale Evans, 'Talking to the Bug," Microcad News, pp. 58-61, Mar. 1989.

Primary Examiner—David R. Hudspeth
Assistant Examiner—Talivaldis Ivars Smits
Attorney, Agent, or Firm—Fish & Richardson PC

[57] ABSTRACT

Words are added to a speech recognition system vocabulary during user dictation by (a) extracting, from a multi-word user utterance, speech frames that correspond to each one of the one or more new words; and (b) building speech models for the one or more new words using the extracted speech frames.

13 Claims, 7 Drawing Sheets