Data processing: speech signal processing – linguistics – language – Speech signal processing – Recognition
Reexamination Certificate
1998-10-23
2001-06-05
Korzuch, William R. (Department: 2641)
Data processing: speech signal processing, linguistics, language
Speech signal processing
Recognition
C704S250000
Reexamination Certificate
active
06243677
ABSTRACT:
TECHNICAL FIELD OF THE INVENTION
This invention relates to a method of out-of-vocabulary word rejection and more particularly to creating speaker-dependent garbage models.
BACKGROUND OF THE INVENTION
Speech recognition systems are now being used in practical applications over telephone channels. Algorithms that are designed for small vocabulary, isolated word applications can produce acceptable recognition rates. A basic assumption for most speech recognition systems is that the utterance to be recognized contains only the words from the recognition vocabulary and background silence. However, previous studies have shown that it is extremely difficult to get real-world users of such systems to speak only the allowable input words. In a large scale trial of a speaker-independent, isolated word technology, it is noted that only 65% of all the callers followed the protocol, although the vocabulary consisted of only five words. Most conventional isolated word speech recognition algorithms are not designed to handle situations where users speak out-of-vocabulary (OOV) words.
In present art systems, there have been studies to solve the OOV problem. An ergodic Hidden Markov Model (HMM), commonly referred to as the “garbage” model, which is trained on speech collected from a large sample of speakers is used to model OOV words. It is generally designed such that each of the states represent a broad phoneme class acoustic, and the transition probabilities represent the frequency of making transitions between pairs of phoneme classes. However, for spoken speed dialing (SSD), which is a speaker-dependent application, there are mainly two problems with this approach. Firstly, since the garbage model is not trained on the particular speaker, its average acoustics may not match the vocal tract characteristics of that person. Secondly, the enrollment may be made under very different conditions (i.e., different levels of background noise, various handset types, etc.) Both of these problems contribute to a mismatch between the within-vocabulary (WV) model set and the garbage model.
SUMMARY OF THE INVENTION
In this invention, we disclose that adapting the garbage model acoustics for each speaker will result in a more accurate model for the OOV words for that speaker.
These and other features of the invention that will be apparent to those skilled in the art from the following detailed description of the invention, taken together with the accompanying drawings.
REFERENCES:
patent: 4802231 (1989-01-01), Davis
patent: 5165095 (1992-11-01), Borcherding
patent: 5325421 (1994-06-01), Hou et al.
patent: 5452397 (1995-09-01), Ittycheriah et al.
patent: 5598507 (1997-01-01), Kimber et al.
patent: 5606643 (1997-02-01), Balasubramanian et al.
patent: 5649057 (1997-07-01), Lee et al.
patent: 5684924 (1997-11-01), Stanley et al.
patent: 5732187 (1998-03-01), Scruggs et al.
patent: 5832429 (1998-11-01), Gammel et al.
patent: 5842165 (1998-11-01), Raman et al.
patent: 5893059 (1999-04-01), Raman
patent: 5895448 (1999-04-01), Vysotsky et al.
C. J. Wellekens, “Mixture density estimators in Viterbi training,” 1992 IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP-92, Mar. 23-26, 1992, vol. 1, pp. 361-364.*
Murakami et al., “Unknown-multiple signal source clustering problem using ergodic HMM and applied to speaker classification,” Fourth International Conference on Spoken Language, ICSLP 96, Oct. 3-6, 1996, vol. 4, pp. 2704-2710.
Arslan Levent M.
Netsch Lorin P.
Rajasekaran Periagaram K.
Korzuch William R.
Lerner Martin
Telecky , Jr. Frederick J.
Texas Instruments Incorporated
Troike Robert L.
LandOfFree
Method of out of vocabulary word rejection does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method of out of vocabulary word rejection, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method of out of vocabulary word rejection will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2493201