Spoken dialog system capable of performing natural interactive access

• US 6,324,513 B1
• Filed: 09/24/1999
• Issued: 11/27/2001
• Est. Priority Date: 06/18/1999
• Status: Expired due to Fees

• Current Assignee: Mitsubishi Denki Kabushiki Kaisha (Mitsubishi Electric Corporation)
• Original Assignee: Mitsubishi Denki Kabushiki Kaisha (Mitsubishi Electric Corporation)
• Inventors: Nagai, Akito, Watanabe, Keisuke, Ishikawa, Yasushi
• Primary Examiner(s): Korzuch, William
• Assistant Examiner(s): ABEBE, DANIEL DEMELASH
• Application Number:
  US09/404,302
• Time in Patent Office: 795 Days
• Field of Search: 704/246, 704/251, 704/256, 704/270, 704/272, 704/275
• US Class Current: 704/275
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue

Abstract

There is provided a spoken dialog system, in which an interactive operation is effectively carried out in a natural manner as to a speech containing words out of a set vocabulary. The spoken dialog system is arranged by comprising: an input storing unit for storing therein an input voice; a dictionary storing unit for classifying a key word to be recognized every item so as to define the classified key word; a dictionary selecting unit for selecting a key word from the dictionary storing unit to thereby output the selected key word as a recognized dictionary; a voice recognizing unit for recognizing the input voice with employment of the recognized dictionary selected from the dictionary selecting unit, and for outputting the extracted key word candidate in combination with the recognition section; a confirmed history storing unit for storing therein a key word candidate confirmed by a user as a determined key word; and a dialog managing unit for judging as to whether or not an input voice other than the determined key word is again recognized and for executing such operations that when the judgment result is the recognition to be once again performed, both an item and the determined key word are transferred as dictionary selection information to the dictionary selecting unit, and a section of an input voice other than the determined key word is updated as a re-recognition section. The updated confirmation section is also instructed to the voice recognizing unit.

Claims

1. A spoken dialog system comprising:

an input storing unit for storing therein an input voice in combination with time information;

a dictionary storing unit for classifying a key word to be recognized every item so as to define the classified key word;

a dictionary selecting unit for selecting a key word corresponding to designated dictionary selection information from said dictionary storing unit to thereby output the selected key word as a recognized dictionary;

a voice recognizing unit for recognizing an input voice or the input voice in said input storing unit with respect to a designated recognition section with employment of the recognized dictionary selected from said dictionary selecting unit, and for outputting the extracted key word candidate in combination with the recognition section;

a confirmed history storing unit for storing therein a history of a determined key word every speech; and

a dialog managing unit for causing a user to confirm the key word candidate outputted from said voice recognizing unit;

for registering the confirmed key word candidate as a determined key word into said confirmed history storing unit;

for judging as to whether or not an input voice other than the determined key word is recognized based upon both a section of the determined key word and a recognition section thereof; and

for executing such operations that when the judgment result is the recognition to be once again, both an item and the determined key word are transferred as dictionary selection information to said dictionary selecting unit;

a section of an input voice other than the determined key word is updated as a re-recognition section; and

also the updated confirmation section is instructed to said voice recognizing unit, whereas when the judgment result is not the recognition to be once again performed, a predetermined interactive operation is carried out in response to the determined key word.

2. A spoken dialog system as claimed in claim 1, further comprising:

a key word position evaluated value storing unit for storing therein an evaluated value with respect to a key word extracted position; and

a candidate selecting unit provided between said voice recognizing unit and said dialog managing unit, for referring to an evaluated value with respect to a position of a key word candidate derived from said voice recognizing unit from said key word position evaluated value storing unit, and for selecting a key word candidate to be confirmed by a user from a plurality of key word candidates corresponding to the confirmed result to thereby send the selected keyword candidate to said dialog managing unit.

3. A spoken dialog system as claimed in claim 1, further comprising:

a functional word storing unit for storing therein a word other than a key word as a functional word in combination with language information;

a connection knowledge storing unit for storing therein a connection relationship between an item and a functional word; and

a functional word extracting unit provided between said input storing unit and said dialog managing unit, for referring to said functional word storing unit so as to extract a functional word candidate from the input voice stored in the input storing unit, and for sending the extracted functional word to said dialog managing unit,

wherein said dialog managing unit selects a functional word candidate connectable to a determined key word by referring to the connection relationship of said connection knowledge storing unit; and

updates a recognition section in a recognition once again performed by referring to time information of said functional word candidate.

4. A spoken dialog system as claimed in claim 3, further comprising:

a functional word position evaluated value storing unit for storing therein an evaluated value at a position of a functional word,

wherein said functional word extracting unit selects a functional word candidate by referring to the evaluated value in said functional word position evaluated value storing unit.

5. A spoken dialog system as claimed in claim 3, further comprising:

an item existence evaluated value storing unit for storing therein an evaluated value with respect to existence of items located before/after a functional word,

wherein said dialog managing unit updates a recognition section in a recognition once again performed by referring to both a functional word candidate in a voice section other than a section occupied by a determined key word, and an item existence evaluated value in said item existence evaluated value storing unit, which corresponds to said functional word candidate.

6. A spoken dialog system as claimed in claim 1, further comprising:

an item relationship rule storing unit for storing therein a cooccurrence relationship among items as an item relationship rule,

wherein said dictionary selecting unit selects an item capable of satisfying the item relationship rule in said item relationship rule storing unit among not-yet-confirmed items as an item to be again recognized; and

selects a key word corresponding to both the item to be again recognized and a determined key word from said dictionary storing unit to thereby provide the selected key word as a recognized dictionary to said voice recognizing unit.

7. A spoken dialog system as claimed in claim 1, further comprising:

an item chain evaluated value storing unit for storing therein an evaluated value with respect to an item chain,

wherein said dictionary selecting unit employs the item chain evaluated value in said item chain evaluated value storing unit as a reference used to select an item to be again recognized among not-yet-confirmed items so as to select a key word corresponding to both the item to be again recognized and the determined key word from said dictionary storing unit.

8. A spoken dialog system as claimed in claim 3, further comprising:

an item/functional word chain evaluated value storing unit for storing therein an evaluated value of an item chain with respect to a set of an item and a functional word, and wherein;

said dialog managing unit sends a functional word candidate connectable to a determined key word to said dictionary selecting unit; and

said dictionary selecting unit employs the item/functional word chain evaluated value in said item/functional word chain evaluated value storing unit as a reference used to select an item to be again recognized among not-yet-confirmed items, so as to select a key word corresponding to both the item to be again recognized and the determined key word from said dictionary storing unit.

9. A spoken dialog system as claimed in claim 7, wherein:

said dictionary selecting unit notifies the item of the determined key word and the item to be again recognized to said voice recognizing unit; and

said voice recognizing unit evaluates an item chain corresponding to a series of key word candidates equal to a recognition candidate by referring to the item chain evaluated value in said item chain evaluated value storing unit.

10. A spoken dialog system as claimed in claim 8, wherein:

said dictionary selecting unit notifies both the item of the determined key word and the item to be again recognized to said voice recognizing unit;

said functional word extracting unit sends the functional word candidate extracted from the input voice in said input storing unit to said voice recognizing unit; and

said voice recognizing unit evaluates an item chain corresponding to a series of key word candidates equal to a recognized candidate by referring to both the item/functional word chain evaluated value in said item/functional word chain evaluated value and also the functional word candidate extracted from said functional word extracting unit.

11. A spoken dialog system as claimed in claim 3, further comprising:

a functional word cooccurrence evaluated value storing unit for storing therein an evaluated value of a cooccurrence relationship between functional words,

wherein said voice recognizing unit enters the functional word candidate from said functional word extracting unit extracted from the input voice in said input storing unit; and

also evaluates a key word candidate equal to a recognized candidate and a series of functional word candidates by referring to the functional word cooccurrence evaluated value in said functional word cooccurrence value storing unit.

12. A spoken dialog system as claimed in claim 1, further comprising:

an acoustic model storing unit for storing therein an acoustic parameter in correspondence with a language unit; and

a speaker adaptive unit for reading out from said input storing unit an input voice corresponding to a section of a determined key word by referring to both a language expression and a section of a determined key word entered from said dialog managing unit, and for learning a parameter of an acoustic model by employing the input voice and the language expression of the determined key word to thereby update the learned parameter of said acoustic model,

wherein said voice recognizing unit recognizes an input voice by employing the updated parameter of the acoustic model derived from said speaker adaptive unit.