Data processing: speech signal processing – linguistics – language – Speech signal processing – Recognition
Reexamination Certificate
1997-11-20
2001-03-27
Hudspeth, David R. (Department: 2741)
Data processing: speech signal processing, linguistics, language
Speech signal processing
Recognition
C704S251000, C704S270000, C455S563000
Reexamination Certificate
active
06208965
ABSTRACT:
BACKGROUND INFORMATION
The present invention is directed to a method and apparatus for recognizing an input identifier, and, in particular, to a method and apparatus that recognizes the input identifier on the basis of a set of comparison identifiers that is generated from the input identifier provided by the user.
Most institutions, such as banks, department stores, and airlines, allow customers to access over the telephone a wide variety of services and information. Before the advent of touch-tone telephones, a customer would obtain these services and information through interacting with a live operator. As touch-tone telephones became more prevalent in homes, these institutions began switching to automated customer-access systems. After dialing a telephone number, a customer using such systems would be asked to enter an account number, or some other identifier. The user would enter the account number through the keys of a touch-tone keypad. As used herein, the term “identifier” refers to a string of characters that may comprise a plurality of letters, numbers, or both; based on this definition, an identifier may comprise either a random aggregation of characters, an actual word, or the name of a place or a person. Furthermore, as used herein, an identifier may comprise, but is not limited to, a string of characters for identifying a particular product, service, location, or account information.
Today, a user can provide such an identifier to an institution through a number of ways besides touch-tone telephony, including by speaking the identifier into the telephone handset. When the user is prompted to speak the identifier, the institution uses a speech recognition system to produce a recognized identifier based on the spoken identifier. Regardless of how the user enters the identifier, however, once a recognized signal representative of the input identifier has been received at the institution, the institution must match the recognized signal with one of a plurality of pre-stored identifiers, which are also referred to as reference identifiers. After a match between the input identifier and a reference identifier is found, the institution performs a transaction based on the matched reference identifier, such as retrieving account information or ordering a specific product. Finding such a match may take some time, however, since institutions often maintain thousands, if not millions, of reference identifiers. Moreover, the more reference identifiers that an institution must search through in order to find a match with an input identifier, the less accurate this search becomes, especially when the reference identifiers are alphanumeric, which tend to be confused with one another. Matching input identifiers with the wrong reference identifiers at least leads to delays and customer dissatisfaction, and perhaps also to erroneous billing or even to divulging sensitive, confidential information. What is therefore needed is a system that maintains a high degree of accuracy in finding a match for an input identifier by carrying out a procedure that eliminates the need to rely on a pre-stored set of reference identifiers.
SUMMARY OF THE INVENTION
The present invention is directed to a method and apparatus that recognizes an input identifier by matching it with one of a plurality of comparison identifiers generated on the basis of the input identifier itself. Thus, the present invention does not use a pre-stored set of reference identifiers to find a match for the input identifier. The present invention first requires a user to provide the input identifier according to a first form. For example, the present invention may require the user to spell a name that is to serve as the input identifier. The present invention provides a first recognized identifier based on the input identifier. On the basis of this recognized identifier, the present invention generates a plurality of comparison identifiers. The present invention then prompts the user to provide the input identifier again, but this time in a different form than the first. For example, the present invention may require the user to pronounce the name that he previously spelled. The present invention then generates a second recognized identifier on the basis of the second entry of the input identifier.
In order to generate the set of comparison identifiers, the present invention uses a confusion matrix, which is a data structure from which the present invention can determine the probability that a particular character of the input identifier provided by the user will be recognized as another character. From this confusion matrix, the present invention determines for each character of the first recognized identifier an alternative character set. The characters of each alternative character set is associated with at least a certain probability of having been spoken as an input letter for the character occupying the associated character position. Thus, if the first character of a recognized identifier is “A”, and it is associated with the alternative character set “A”, “J”, and “K”, what this means is that there was at least a certain probability that the user spoke either “J” or “K,” but the present invention recognized this pronunciation as an “A”.
As an alternative way to generate the set of comparison identifiers, the present invention employs a collection of pre-stored confusion sets. The characters of each confusion set have a relatively high probability of being confused with the other characters in the same set. Characters from different confusion sets have a relatively low probability of being confused with one another. When using confusion sets, the present invention determines for the first recognized identifier which confusion set includes which characters of the first recognized identifier. Based on these determined confusion sets, the present invention generates the set of comparison identifiers.
In order to take into account the possibility that the character length of the generated comparison identifiers is not equal to the character length of the input identifier, the set of comparison identifiers is expanded by the addition of further comparison identifiers that are of a variable character length. Each of these variable-length comparison identifiers are determined in accordance with pre-stored rules that allow for certain character additions, substitutions, or deletions. In order to compare the second identifiers only with valid comparison identifiers, the present invention eliminates from the set of comparison identifiers those comparison identifiers that comprise mere aggregation of characters of no intelligible arrangement. For example, if the identifiers correspond to actual words, the present invention employs a set of pre-stored spelling rules to root out those identifiers that do not have accepted English-language spellings. Thus, the present invention may be programmed to eliminate those comparison identifiers that do not follow the letter “Q” with the letter “U”.
After the set of comparison identifiers has been generated according to the above-discussed techniques, the present invention determines whether the second recognized identifier matches any of the comparison identifiers. If a match is found, then the matched comparison identifier is selected as corresponding to the input identifier. It should be appreciated that the techniques involving confusion matrices and confusion sets are but two ways by which a plurality of comparison identifiers may be generated based on an input identifier, and that the present invention encompasses other techniques for generating comparison identifiers from an input identifier.
REFERENCES:
patent: 4783804 (1988-11-01), Juang et al.
patent: 4817156 (1989-03-01), Bahl et al.
patent: 4819271 (1989-04-01), Bahl et al.
patent: 5050215 (1991-09-01), Nishimura
patent: 5125022 (1992-06-01), Hunt et al.
patent: 5127043 (1992-06-01), Hunt et al.
patent: 5216720 (1993-06-01), Naik et al.
patent: 5255310 (1993-10-01), Kim et al.
patent: 5297194 (1994-03-01), Hunt et al.
patent: 53
Brown Deborah W.
Goldberg Randy G.
Rosinski Richard R.
AT&T Corp.
Hudspeth David R.
Kenyon & Kenyon
Wieland Susan
LandOfFree
Method and apparatus for performing a name acquisition based... 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 apparatus for performing a name acquisition based..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for performing a name acquisition based... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2480353