Method of interpretation service for voice on the phone
View Patent ↗An interpretation service for voice based on sentence template retrieval allows a translation database to be customized without burdening users and enables sentences needed by users to be accurately interpreted. A sentence to be stored in a translation database for customization can be described as a sentence template including a slot which allows words to be replaced. A condition for selecting sentence templates is extracted from a registered user profile (UP). A sentence template matching the condition is retrieved from those stored in the translation database for customization and is registered in a translation database customized for each user. A word extracted from the UP is inserted into the sentence template's slot for registration to a sentence dictionary customized for each user.
1. A computer implemented method of customizing a dictionary used for speech recognition, comprising the steps of:
extracting user-specific vocabulary from a set of registered user profile information; and
generating a new customized sentence by inserting at least one word in said user-specific vocabulary into a slot of a sentence template stored in a translation database and storing said customized sentence into said dictionary, wherein said customized sentence stored in said dictionary is used for recognizing speech input by a user that is converted into voice data and collated with said customized sentence.
2. A computer implemented method of customizing a dictionary according to claim 1 , wherein said set of registered user profile information is generated based on inputted answers to a questionnaire for said user.
3. A computer implemented method of customizing a dictionary according to claim 1 , wherein said sentence template includes slot definition information including a category of words that can be inserted in said slot, and said at least one word in said user-specific vocabulary is inserted in said slot according to said slot definition information.
4. A method of customizing a dictionary according to claim 1 , wherein said step of storing said customized sentence into said dictionary includes storing said customized sentence as a time series of feature vectors.
5. A method of customizing a dictionary according to claim 1 , wherein said step of storing said customized sentence into said dictionary includes storing said customized sentence as a time series of feature vectors and said step of recognizing the input speech by collating said corresponding voice data with said customized sentence includes collating values of a time series of feature vectors of the input speech with the stored time series of feature vectors of the customized sentences in the sentence dictionary.
6. A storage media readable by a computer for storing a customizing program for customizing a dictionary used for speech recognition, said customizing program enabling the computer to perform the steps of:
extracting user-specific vocabulary from a set of registered user profile information; and
generating a new customized sentence by inserting at least one word in said user-specific vocabulary in a slot of a sentence template stored in a translation database and storing said customized sentence into said dictionary, wherein said customized sentence stored in said dictionary is used for recognizing speech input by a user that is converted into voice data and collated with said customized sentence.
7. A storage media readable by a computer for storing a customizing program for customizing a dictionary used for speech recognition according to claim 6 , wherein said step of storing said customized sentence into said dictionary includes storing said customized sentence as a time series of feature vectors.
8. A storage media readable by a computer for storing a customizing program for customizing a dictionary used for speech recognition according to claim 6 , wherein said step of storing said customized sentence into said dictionary includes storing said customized sentence as a time series of feature vectors and said step of recognizing the input speech by collating said corresponding voice data with said customized sentence includes collating values of a time series of feature vectors of the input speech with the stored time series of feature vectors of the customized sentences in the sentence dictionary.