Data processing: speech signal processing – linguistics – language – Speech signal processing – Recognition
Reexamination Certificate
2000-11-13
2004-10-05
Chawan, Vijay (Department: 2654)
Data processing: speech signal processing, linguistics, language
Speech signal processing
Recognition
C704S256000, C704S251000, C704S231000, C704S270000
Reexamination Certificate
active
06801890
ABSTRACT:
CROSS-REFERENCE TO RELATED APPLICATIONS
This application is entitled to the benefit of and incorporates by reference in their entireties essential subject matter disclosed in International Application No. PCT/DE99/00137, filed on Jan. 20, 1999 and German Patent Application No. 19894047.4, filed on Feb. 3, 1998.
BACKGROUND OF THE INVENTION
The invention relates to a method for enhancing the recognition probability in voice recognition systems.
1. Field of the Invention
2. Description of the Related Art
U.S. Pat. No. 5,617,468 discloses a method for enhancing the recognition probability in voice recognition systems wherein, after input of a term to be recognized, a post-training of the previously stored homonymic term by means of the input term is carried out. This method does not permit reliable voice recognition in acoustically changing environments.
EP Patent No. 0 241 163 relates to a voice recognition method that alerts the user if a term to be recognized is already stored in a similar form in the voice recognition system and the two terms might be confused. This method also does not offer a reliable recognition of terms in changing acoustic environments.
BRIEF SUMMARY OF THE INVENTION
Voice recognition systems today are used primarily in computers, communication systems and other technical equipment where ease of operation or fast data input is important. The prior art systems, however, are not mature and are flawed in operation, particularly if they are operated in environments with acoustic interference. In this case, a word to be recognized is often misrecognized or not recognized at all. As a result, the user must multiply repeat the word to be recognized, which causes unreasonable delays if recognition errors occur frequently.
U.S. Pat. No. 5,617,468 discloses a method for enhancing the recognition probability in voice recognition systems wherein, after input of a term to be recognized, a post-training of the previously stored homonymic term by means of the input term is carried out. This method does not permit reliable voice recognition in acoustically changing environments.
EP-A-0 241 163 relates to a voice recognition method that alerts the user if a term to be recognized is already stored in a similar form in the voice recognition system and the two terms might be confused. This method also does not offer a reliable recognition of terms in changing acoustic environments.
Thus, the object of the present invention is further to develop to propose a method for voice recognition that enhances the speaker-dependent recognition rate in a user-friendly manner, particularly in environments with acoustic interference.
This object is attained by the characteristic features of claim
1
.
Advantageous modifications and further developments of the invention are set forth in the dependent claims.
The invention proposes that in voice recognition systems a post-training of the voice patterns of newly input terms. rate, particularly in environments with acoustic interference, is enhanced.
The invention proposes that in voice recognition systems a post-training of the voice patterns of newly input terms, which may have been misrecognized or not recognized at all, be carried out. Post-training means that a misrecognized or non-recognized term is not simply overwritten after repeated input, but is compared and correlated with the previously input terms or supplemented by a new pattern in order to reduce or mask out patterns or incidental noise that are unimportant for recognizing the term. The invention is intended, in particular, for use in voice recognition systems that operate in environments with acoustic interference, e.g. in mobile radiocommunications terminals, telephones, etc.
If a term previously stored as a voice pattern is successfully recognized, the stored voice pattern is post-trained with the newly recorded pattern. In case of successful recognition of each previously recorded term, this post-training is carried out n times, where the number of passes can be freely selected and changed at any time. This post-training makes it possible continuously to reduce the influence of variable incidental noise on the actual constant voice pattern of a word.
According to a further development of the invention, if the recognition of a word is uncertain, the system prompts for a renewed input, which is used for post-training. Here, too, the maximum number of passes can be freely selected and changed at any time. If the comparison of an input voice pattern with stored voice patterns results in a similar and little differentiated recognition probability for several stored terms, the system plays these terms for the user and prompts the user to repeat the initially input term. If recognition is then successful, post-training is interrupted. For reasons of clarity, it is preferred to limit the number of the possible terms that are output by the system in case of uncertain recognition to a predefined number, and to limit the repetition of the process to, e.g. three direct repetitions.
Another further development provides that in case of a new input of a word or term in the system, the system carries out a comparison with previously stored terms after the new term has been input. The voice recognition system is thus used to compare a new term with previously stored terms and to determine whether the voice pattern of the new term is sufficiently distinct from the voice patterns of the previously stored terms that a misrecognition or uncertain recognition is not expected. However, if in the context of a correlation comparison based on a defined criterion of uncertain recognition, the new voice pattern results in a strong similarity or probable match with previously stored voice patterns, the system optically or acoustically informs the user and prompts him to repeat the input of the new term, which serves for post-training. The number of consecutive repetions can be freely selected and changed at any time.
REFERENCES:
patent: 4489434 (1984-12-01), Moshier
patent: 4489435 (1984-12-01), Moshier
patent: 4672668 (1987-06-01), Kitazume et al.
patent: 5040213 (1991-08-01), Yasuda et al.
patent: 5329609 (1994-07-01), Sanada et al.
patent: 5617486 (1997-04-01), Chow et al.
patent: 5864803 (1999-01-01), Nussbaum
patent: 6044343 (2000-03-01), Cong et al.
patent: 6178404 (2001-01-01), Hambleton et al.
patent: 6347297 (2002-02-01), Asghar et al.
Kauschke Ulrich
Rast Herbert Roland
Runge Fred
Chawan Vijay
DeTeMobil Deutsche Telekom MobilNet GmbH
McCormick Paulding & Huber LLP
LandOfFree
Method for enhancing recognition probability in voice... 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 for enhancing recognition probability in voice..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for enhancing recognition probability in voice... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3329674