Speech recognition apparatus, speech recognition method, conversation control apparatus, conversation control method, and programs for therefor

• US 20050021331A1
• Filed: 06/17/2004
• Published: 01/27/2005
• Est. Priority Date: 06/20/2003
• Status: Active Grant

• Current Assignee: MyClerk LLC (IPValuation Partners, LLC)
• Original Assignee: P B Limited (Balfour Beatty Plc)
• Inventors: Katukura, Hiroshi, Huang, Shengyang
• Granted Patent: US 7,415,406 B2
• Time in Patent Office: Days
• US Class Current: 704/231
• CPC Class Codes: G10L 15/08 Speech classification or search
• CPC Class Codes: G10L 15/1815 Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning

Abstract

An automatic conversation apparatus includes a speech recognizing unit receiving a speech signal and outputting characters/character string corresponding to the speech signal as a recognition result; a speech recognition dictionary storing unit storing a language model for determining candidates corresponding to the speech signal; a conversation database storing plural pieces of topic specifying information; a sentence analyzing unit analyzing the characters/character string outputted from the speech recognizing unit; and a conversation control unit storing a conversation history and acquiring an answer sentence based on an analysis result of the sentence analyzing unit. Speech recognizing unit includes a collating unit that outputs plural candidates based on the speech recognition dictionary storing unit; and a candidate determining unit comparing the plural candidates outputted from collating unit with topic specifying information corresponding to the conversation history with reference to the conversation database and outputs one candidate based on the comparison.

Claims

1. A speech recognition apparatus comprising:

a speech recognizing unit;

a speech recognition dictionary storing unit;

a conversation database that has stored therein plural pieces of topic specifying information; and

a conversation history storing memory that stores subjects included in conversation in the past, wherein;

said speech recognizing unit selects plural candidates corresponding to a speech signal on the basis of contents stored in the speech recognition dictionary storing unit, selects a candidate from the selected plural candidates on the basis of topic specifying information associated with conversation histories stored in the conversation history storing memory, and outputs the selected candidate as a recognition result.

18. A program for causing a computer to function as the speech recognition apparatus according to claim 1.

2. A speech recognition apparatus comprising:

a speech recognizing unit that receives a speech signal;

a speech recognition dictionary storing unit that has stored therein a language model for determining candidates corresponding to said speech signal;

a conversation database that has stored therein plural pieces of topic specifying information; and

a conversation history storing memory that stores a conversation history, wherein;

the speech recognizing unit has;

a word collating unit that outputs plural candidates with reference to said speech recognition dictionary storing unit on the basis of the inputted speech signal; and

a candidate determining unit that compares the plural candidates outputted from the word collating unit with topic specifying information corresponding to the conversation history with reference to said conversation database and outputs any one candidate as a recognition result on the basis of a result of the comparison.

3. The speech recognition apparatus according to claim 2, wherein:

said conversation database has stored therein first topic specifying information relating to subjects of conversation and second topic specifying information not relating to the subjects of conversation, and

said candidate determining unit compares said plural candidates and the first topic specifying information corresponding to a conversation history and outputs any one of said plural candidates as a recognition result.

4. The speech recognition apparatus according to claim 2, wherein:

said candidate determining unit compares said plural candidates and topic specifying information corresponding to a conversation history with reference to said conversation database, judges whether there is a candidate, which matches the topic specifying information corresponding to the conversation history, in said plural candidates, and if there is a candidate matching the topic specifying information, outputs the candidate as a recognition result.

5. The speech recognition apparatus according to claim 4, wherein:

said conversation database has stored therein first topic specifying information relating to subjects of conversation and second topic specifying information not relating to the subjects of conversation, and

said candidate determining unit compares said plural candidates with the first topic specifying information corresponding to a conversation history and outputs any one of said plural candidates as a recognition result.

6. The speech recognition apparatus according to claim 4, wherein:

the speech recognition apparatus further includes a candidates narrowing-down unit, and

wherein, as a result of comparing said plural candidates and the topic specifying information corresponding to the conversation history, if there is no candidate matching the topic specifying information, said candidate determining unit requests said candidates narrowing-down unit to narrow down the candidates, and said candidates narrowing-down unit outputs a candidate, which has a maximum likelihood, among said plural candidates as a recognition result in response to this request.

7. The speech recognition apparatus according to claim 6, wherein:

said conversation database has stored therein first topic specifying information relating to subjects of conversation and second topic specifying information not relating to the subjects of conversation, and

said candidate determining unit compares said plural candidates and the first topic specifying information corresponding to a conversation history and outputs any one of said plural candidates as a recognition result.

19. A program for causing a computer to function as the speech recognition apparatus according to claim 2.

8. A speech recognition method for outputting characters or a character string corresponding to an inputted speech as a recognition result, comprises the steps of:

analyzing acoustic characteristics on the basis of a speech signal obtained from the inputted speech;

outputting plural candidates on the basis of said acoustic characteristics;

acquiring topic specifying information corresponding to a conversation history stored in advance; and

comparing said plural candidates and the acquired topic specifying information and outputs any one of said plural candidates as a recognition result.

9. The speech recognition method according to claim 8, wherein:

said topic specifying information includes first topic specifying information relating to subjects of conversation and second topic specifying information not relating to the subjects of conversation, and

the topic specifying information, which is compared with said plural candidate in the step of outputting a candidate as a recognition result, is limited to the first topic specifying information.

10. The speech recognition method according to claim 8, wherein:

in the step of outputting a candidate as a recognition result, it is judged whether there is a candidate, which matches the topic specifying information corresponding to a conversation history, in said plural candidates, and if there is a candidate matching the topic specifying information, the candidate is outputted as the recognition result.

11. The speech recognition method according to claim 10, wherein:

said topic specifying information includes first topic specifying information relating to subjects of conversation and second topic specifying information not relating to the subjects of conversation, and

the topic specifying information, which is compared with said plural candidate in the step of outputting a candidate as a recognition result, is limited to the first topic specifying information.

12. The speech recognition method according to claim 10, wherein:

as a result of comparing said plural candidates and the topic specifying information corresponding to the conversation history in the step of outputting a candidate as a recognition result, if there is no candidate matching the topic specifying information, a candidate having a maximum likelihood among said plural candidates is outputted as the recognition result.

13. The speech recognition method according to claim 12, wherein:

said topic specifying information includes first topic specifying information relating to subjects of conversation and second topic specifying information not relating to the subjects of conversation, and

the topic specifying information, which is compared with said plural candidate in the step of outputting a candidate as a recognition result, is limited to the first topic specifying information.

20. A program for causing a computer to execute the speech recognition method according to claim 8.

14. An automatic conversation apparatus comprising:

a speech recognizing unit that receives a speech signal and outputs characters or a character string corresponding to the speech signal as a recognition result;

a speech recognition dictionary storing unit that has stored therein a language model for determining candidates corresponding to the speech signal;

a conversation database that has stored therein plural pieces of topic specifying information;

a sentence analyzing unit that analyzes the characters or the character string outputted from said speech recognizing unit; and

a conversation control unit that stores a conversation history and acquires an answer sentence on the basis of an analysis result of said sentence analyzing unit, wherein;

said speech recognizing unit includes;

a word collating unit that outputs plural candidates with reference to said speech recognition dictionary storing unit on the basis of the inputted speech signal; and

a candidate determining unit that compares the plural candidates outputted from said word collating unit with topic specifying information corresponding to the conversation history with reference to said conversation database and outputs any one candidate as a recognition result on the basis of a result of the comparison.

15. The automatic conversation apparatus according to claim 14, wherein:

said conversation database has stored therein first topic specifying information relating to subjects of conversation and second topic specifying information not relating to the subjects of conversation, and

said candidate determining unit compares said plural candidates and the first topic specifying information corresponding to a conversation history and outputs any one of said plural candidates as a recognition result.

21. A program for causing a computer to function as the automatic conversation apparatus according to claim 14.

16. An automatic conversation method for outputting a response sentence in response to an input speech, comprising the steps of:

analyzing acoustic characteristics on the basis of a speech signal obtained from the inputted speech;

outputting plural candidates on the basis of said acoustic characteristics;

acquiring topic specifying information corresponding to a conversation history stored in advance;

comparing said plural candidates and the acquired topic specifying information and outputs any one of said plural candidates as a recognition result; and

analyzing said recognition result and outputs an answer sentence associated with the topic specifying information corresponding to said conversation history on the basis of said recognition result.

17. The automatic conversation method according to claim 16, wherein:

said topic specifying information includes first topic specifying information relating to subjects of conversation and second topic specifying information not relating to the subjects of conversation, and

the topic specifying information, which is compared with said plural candidate in the step of outputting a candidate as a recognition result, is limited to the first topic specifying information.

22. A program for causing a computer to execute the automatic conversation method according to claim 16.

23. A speech recognition apparatus comprising:

speech recognizing means;

speech recognition dictionary storing means;

conversation database means that has stored therein plural pieces of topic specifying information; and

conversation history storing means that stores subjects included in conversation in the past, wherein;

said speech recognizing means selects plural candidates corresponding to a speech signal on the basis of contents stored in the speech recognition dictionary storing means, selects a candidate corresponding to topic specifying information, which is associated with a conversation history stored in said conversation history storing means, from the selected plural candidates, and outputs the candidate as a recognition result.

24. A speech recognition apparatus comprising:

speech recognizing means that receives a speech signal;

speech recognition dictionary storing means that has stored therein a language model for determining candidates corresponding to said speech signal;

conversation database means that has stored therein plural pieces of topic specifying information; and

conversation history storing means that stores a conversation history, wherein;

the speech recognizing means has;

collating means that outputs plural candidates with reference to said speech recognition dictionary storing means on the basis of the inputted speech signal; and

candidate determining means that compares the plural candidates outputted from the collating means with topic specifying information corresponding to the conversation history with reference to said conversation database means and outputs any one candidate as a recognition result on the basis of a result of the comparison.

25. An automatic conversation apparatus comprising:

speech recognizing means that receives a speech signal and outputs characters or a character string corresponding to the speech signal as a recognition result;

speech recognition dictionary storing means that has stored therein a language model for determining candidates corresponding to the speech signal;

conversation database means that has stored therein plural pieces of topic specifying information;

sentence analyzing means that analyzes the characters or the character string outputted from said speech recognizing means; and

conversation control means that stores a conversation history and acquires an answer sentence on the basis of an analysis result of said sentence analyzing means, wherein;

said speech recognizing means has;

collating means that outputs plural candidates with reference to said speech recognition dictionary storing means on the basis of the inputted speech signal; and

candidate determining means that compares the plural candidates outputted from said collating means with topic specifying information corresponding to the conversation history with reference to said conversation database means and outputs any one candidate as a recognition result on the basis of a result of the comparison.