Data processing: speech signal processing – linguistics – language – Speech signal processing – Recognition
Reexamination Certificate
1999-05-14
2001-10-23
Dorvil, Richemond (Department: 2741)
Data processing: speech signal processing, linguistics, language
Speech signal processing
Recognition
C704S255000
Reexamination Certificate
active
06308151
ABSTRACT:
CROSS REFERENCE TO RELATED APPLICATIONS
Not Applicable
STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT
Not Applicable
BACKGROUND OF THE INVENTION
1. Technical Field
This invention relates to the field of speech recognition computer applications and more specifically to a system for automatically updating a speech recognition system's vocabulary and language model with new words and context information, respectively, extracted from text of an incoming E-mail message in order to improve dictation accuracy of the E-mail response.
2. Description of the Related Art
Speech recognition is the process by which acoustic signals received by microphone are converted to a set of words by a computer. These recognized words may then be used in a variety of computer software applications for purposes such as document preparation, data entry, and command and control. Speech recognition is generally a difficult problem due to the wide variety of pronunciations, individual accents and speech characteristics of individual speakers. Consequently, language models are often used to help reduce the search space of possible words and to resolve ambiguities as between similar sounding words. Language models tend to be statistically based systems and can be provided in a variety of forms. The simplest language model can be specified as a finite state network, where the permissible words following each word are given explicitly. However, more sophisticated language models have also been developed which are specified in terms of probabilities of word sequences.
Conventional speech recognition systems permit language models to be updated by analyzing samples of existing text. The analysis process in such conventional systems involves a process whereby the speech recognition software compiles statistics relating to the likelihood that a particular word precedes or follows some other word. A bigram model or sometimes a trigram is typically used to represent this data with regards to certain word pairs or even triplets. The analysis process is typically initiated by conventional systems when a new user is established for the system, or when a user manually initiates the analysis process.
It would be desirable to provide a method of allowing a speech recognition system to automatically determine whether to update the language model using a particular existing text. For similar reasons, when the speech recognition system determines appropriate the use of a particular existing text in updating the language model, it would also be desirable for the system to update the system vocabulary to include new words from the subject text.
Such a method would be particularly desirable in the context of using a speech recognition system to dictate an E-mail response to an incoming E-mail message. The reason for this stems from the fact that: 1) an E-mail response typically involves the same subject matter as the incoming E-mail message; and 2) most E-mail authors compose messages which include creative, informal, or unusual words and phrases. Conventional speech recognition vocabularies and language models simply do not account for such atypical language, as the vocabularies and models have finite capacity incapable of including the wide variety of atypical words and phrases that may be used in E-mail messages. For example, the incoming message may contain specific names of individuals, businesses, projects, etc., which are often not included in general purpose vocabularies with a limited vocabulary size. Therefore, the words used in the incoming message and their context are likely to be useful for improving the accuracy of the dictated response.
The present approach consists of dictating the E-mail response, including atypical words borrowed from the prompting E-mail message. The problem is that the system vocabulary and language model will not include the atypical language, and therefore, certain words from the dictated text will be misrecognized by the speech recognition system. Thus, the user must correct the erroneous words using a computer keyboard, an inconvenience users prefer to avoid.
Accordingly, there is a need to provide a system facilitating more efficient dictation of E-mail responses to incoming E-mail messages.
REFERENCES:
patent: 5634086 (1997-05-01), Rtischev et al.
patent: 5675706 (1997-10-01), Lee et al.
ICASSP-95. Gauvain et al. “Developments in continuous speech dictation using the ARPA WSJ task”. May 1995.*
IEEE Asia Pacific Conference Circuits and Systems. Lee, “Spontaneous dialogue recognition with large out-of-vocabularies”. Nov. 1996.*
Proceeding of the 1998 IEEE International Conference on Acoustics Speech and Signal Processing, 1998. Geutner et al., “Adaptive vocabularies for transcribing multilingual broadcast news”. May 1998.
Dorvil Richemond
International Business Machines Corp.
Senterfitt Akerman
LandOfFree
Method and system using a speech recognition system to... 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 and system using a speech recognition system to..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and system using a speech recognition system to... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2553806