Automatic dialog system with database language model

• US 20040034518A1
• Filed: 09/04/2002
• Published: 02/19/2004
• Est. Priority Date: 01/10/2001
• Status: Active Grant

• Current Assignee: Koninklijke Philips Electronics N.V. (Koninklijke Philips N.V.)
• Original Assignee: Koninklijke Philips Electronics N.V. (Koninklijke Philips N.V.)
• Inventors: Te Vrugt, Juergen, Rose, Geog
• Granted Patent: US 7,424,428 B2
• Time in Patent Office: Days
• US Class Current: 704/1
• CPC Class Codes: G10L 15/19 Grammatical context, e.g. disambiguation of the recognition hypotheses based on word sequence rules
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence

Abstract

The invention relates to an automatic dialog system for spoken inquiries into a database entry which contains several components available for inquiry, wherein the speech recognition (3) of a spoken uttrance for inquiring into the database entry is supported by a language model (6) which was prepared before the start of the dialog to which the spoken utterance belongs and which models the correlated occurrence of the components of the database entry provided for the inquiry in the spoken utterance of the dialog.

Claims

1. An automatic dialog system for spoken inquiries into a database entry which contains several components available for inquiry, wherein the speech recognition (3) of a spoken utterance for inquiring into the database entry is supported by a language model (6) which is prepared before the start of the dialog to which the spoken utterance belongs and which models the correlated occurrence of the components of the database entry provided for the inquiry in the spoken utterance of the dialog.

2. An automatic dialog system for spoken database inquiries as claimed in claim 1, characterized in that

the language model (6) assigns the score 1 (“

admitted”

) to the occurrence of several components of a data record in the spoken utterance which is contained in the inquired database, and

the language model (6) assigns the score 0 (“

not admitted”

) to the occurrence of several components of a data record in the spoken utterance which is not contained in the inquired database.

3. An automatic dialog system for spoken database inquiries as claimed in claim 1, characterized in that

the language model (6) assigns to the occurrence of several components of a data record in the spoken utterance which is contained in the inquired database a score in accordance with their relative frequency in the inquired database, and

the language model (6) assigns the score “

0”

(“

not admitted”

) to the occurrence of several components of a data record in the spoken utterance which is not contained in the inquired database.

4. An automatic dialog system for spoken database inquiries as claimed in claim 1, characterized in that

the language model (6) assigns to the occurrence of several components of a data record in the spoken utterance which is contained in the inquired database a score in accordance with their relative frequency in the inquired database, and

the language model (6) assigns to the occurrence of several components of a data record in the spoken utterance which is not contained in the inquired database a score in accordance with the relative frequency of such spoken utterance.

5. A speech recognition system (3) for recognizing a spoken utterance for the spoken inquiry after a database entry which comprises several components available for inquiry, wherein the speech recognition system (3) is supported by a language model (6) which is prepared before the start of the dialog to which the spoken utterance belongs and which models the correlated occurrence of the components of the database entry provided for the inquiry in the spoken utterance of the dialog.

6. A language model (6) for supporting the speech recognition (3) of a spoken utterance for spoken inquiries of a database entry which comprises several components available for inquiry, wherein the language model (6) is prepared before the start of the dialog to which the spoken utterance belongs, and wherein the language model (6) models the correlated occurrence of the components of the database entry provided for the inquiry in the spoken utterance of the dialog.

7. A method of orally inquiring into a database entry which comprises several components available for inquiry, wherein the speech recognition (3) of a spoken utterance for inquiring into the database entry is supported by a language model (6) which was constructed before the start of the dialog to which the spoken utterance belongs and which models the correlated occurrence of the component of the database entry provided for the inquiry in the spoken utterance of the dialog.