Data processing: speech signal processing – linguistics – language – Speech signal processing – Recognition
Reexamination Certificate
1998-08-14
2001-02-20
Hudspeth, David R. (Department: 2741)
Data processing: speech signal processing, linguistics, language
Speech signal processing
Recognition
C704S239000
Reexamination Certificate
active
06192337
ABSTRACT:
BACKGROUND OF THE INVENTION
Developments in speech recognition technology have led to widespread and varied use of speech recognition systems in applications which rely on spoken input words or commands to perform some function. The use of speech recognition techniques in a repertory telephone voice dialer application is one example. It is known that the repertory dialing application allows users to train their own vocabularies for the purpose of associating a phone number to be dialed with each entry in the vocabulary. This can also be applied to other situations when a vocabulary word is trained and the system takes some action when the word is subsequently recognized. However, the list of words often grows to such an extent that it is difficult for an application user to remember when a word has already been entered. Alternatively, a large vocabulary also poses a problem to a user when a word is too similar to another one such that the speech recognizer is much less accurate on these words, if they appeared on the same list.
Traditionally, such systems have attempted to offer the capability to reject such utterances based on comparing the input speech for training the current word to all previously enrolled models. This requires a match that produces often one or more (in systems using N-best outputs) words and, if the resulting word is not the currently trained one or it is a word which has a very poor score, the utterance is added. This technique ignores the models themselves and uses only the correlation between the input speech and the collection of models to do the rejection.
Now, while the traditional systems attempt to handle detecting similar words, these systems cannot handle the case when two or more lists are being combined or more generally the case of manipulating vocabularies when the input audio is no longer available.
SUMMARY OF THE INVENTION
It is to be appreciated that the present invention applies to the rejection not only of homonyms (acoustically similar words) but to the more general category of acoustically similar sounds known as homophones. Accordingly, it is to be understood that the term homophone, as referred to herein, includes acoustically similar single and multiple phone words as well as individual phones themselves, whereby the words or phones may have meanings and/or no meanings at all.
The present invention provides apparatus and methods to reject acoustically trained words by comparing the set of models to determine if any words in the vocabulary are homophones. If so, then the word is rejected and not added to the vocabulary.
The method preferably involves taking, as input, the set of models to be checked and doing a distance metric on the models to produce a score and subsequently comparing this score with a threshold, and those words which fall under this threshold are declared to be homophones and rejected.
In a repertory dialing application, a user is allowed to add names to the system. When the list size is quite large, it's often possible that the user will try to enter either a name that sounds too close to another name on the list, such that recognition accuracy will suffer, or may try to enter a duplicate name. The present invention provides apparatus and methods which compare the models directly to see when phrases are too similar.
In one aspect of the invention, a method of training at least one new word for addition to a vocabulary of a speech recognition engine containing existing words comprises the steps of: a user uttering the at least one new word; computing respective measures between the at least one newly uttered word and at least a portion of the existing vocabulary words, the respective measures indicative of acoustic similarity between the at least one word and the at least a portion of existing words; if no measure is within the threshold range, automatically adding the at least one newly uttered word to the vocabulary; and if at least one measure is within a threshold range, refraining from automatically adding the at least one newly uttered word to the vocabulary.
These and other objects, features and advantages of the present invention will become apparent from the following detailed description of illustrative embodiments thereof, which is to be read in connection with the accompanying drawings in which the same reference numerals are used throughout the various figures to designate same or similar components.
REFERENCES:
patent: 4829576 (1989-05-01), Porter
patent: 4918732 (1990-04-01), Gerson et al.
patent: 5033087 (1991-07-01), Bahl et al.
patent: 5218668 (1993-06-01), Higgins et al.
patent: 5349645 (1994-09-01), Zhao
patent: 5621857 (1997-04-01), Cole et al.
patent: 5625748 (1997-04-01), McDonough et al.
patent: 5675704 (1997-10-01), Juang et al.
patent: 5680511 (1997-10-01), Baker et al.
patent: 5715367 (1998-02-01), Gilick et al.
patent: 5752001 (1998-05-01), Dulong
patent: 5778344 (1998-07-01), Attwater et al.
patent: 5850627 (1998-12-01), Gould et al.
patent: 5852801 (1998-12-01), Hon et al.
patent: 5864810 (1999-01-01), Digalakis et al.
patent: 6005549 (1999-12-01), Forest
patent: 6023673 (2000-02-01), Bakis et al.
Ittycheriah Abraham
Maes Stephane H.
F. Chau & Associates LLP
Hudspeth David R.
International Business Machines - Corporation
Wieland Susan
LandOfFree
Apparatus and methods for rejecting confusible words during... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Apparatus and methods for rejecting confusible words during..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Apparatus and methods for rejecting confusible words during... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2559013