Spoken dialogue speech recognition using estimated number of retrieved data as cost

• US 8,364,493 B2
• Filed: 03/11/2009
• Issued: 01/29/2013
• Est. Priority Date: 03/11/2008
• Status: Expired due to Fees

• Current Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Original Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Inventors: Yano, Takehide
• Primary Examiner(s): Smits, Talivaldis Ivars
• Application Number:
  US12/401,640
• Publication Number:
  US 20090234650A1
• Time in Patent Office: 1,420 Days
• Field of Search: 704/251, 704/270, 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
• CPC Class Codes: G10L 2015/228 of application context

Abstract

A spoken dialogue apparatus is provided. The apparatus includes a speech recognition unit, a retrieval unit, a calculation unit, and a selection unit. The speech recognition unit performs speech recognition of an input speech to obtain a recognition candidate. The retrieval unit updates a retrieval condition by use of the recognition candidate and outputs a retrieval result. The calculation unit calculates estimated numbers of data as costs concerning a first response and a second response based on the retrieval condition and the retrieval result. The selection unit selects a response having a lowest cost and presents the response to the user.

Claims

1. A spoken dialogue apparatus which retrieves desired data from a data storage unit storing data in association with attribute values concerning a plurality of attributes of the data, comprising:

a speech recognition unit configured to perform speech recognition of an input speech of a user for a presented response from the apparatus to obtain a recognition candidate in one attribute from recognition target words corresponding to the presented response;

a retrieval unit configured to update a retrieval condition by use of the recognition candidate and retrieve data corresponding to the retrieval condition from the data storage unit to output as a retrieval result;

a calculation unit configured to calculate estimated numbers of data as costs concerning a first response and a second response based on the retrieval condition and the retrieval result, the data being included in retrieval result after narrowed retrieval performed in accordance with an input speech corresponding to the first response asking the user to input an attribute value of an attribute which has not been input and corresponding to the second response ascertaining the truth of an attribute value of an attribute which has been input; and

a selection unit configured to select a response having the lowest cost and present the response to the user,

wherein the retrieval unit (a) updates the retrieval condition by adding the recognition candidate to the retrieval condition when the presented response corresponds to the first response, (b) updates the retrieval condition by maintaining an approved attribute value in the retrieval condition and deleting other attribute value of the attribute related to the presented response from the retrieval condition when the presented response corresponds to the second response and the recognition candidate indicates approval, and (c) updates the retrieval condition by deleting a denied attribute value form the retrieval condition and maintaining other attribute value of the attribute related to the presented response in the retrieval condition when the presented response corresponds to the second response and the recognition candidate indicates denial.

2. The apparatus according to claim 1, wherein the second response ascertains the truth of an attribute value having a highest score provided by the speech recognition unit regarding the attribute related to the presented response.

3. The apparatus according to claim 1, wherein the calculation unit further calculates a cost of a third response ascertaining the truth of an attribute value of the attribute which has not been input.

4. The apparatus according to claim 1, wherein the retrieval unit modifies the retrieval condition by deleting, from the retrieval condition, all the attribute values but an attribute value previously ascertained to be true with regard to the second response when the retrieval result is null.

5. The apparatus according to claim 1, wherein the speech recognition unit again performs speech recognition of a previous input speech in order to obtain a recognition candidate concerning a first unapproved attribute from the recognition target words when an input speech indicating that an attribute value of a second unapproved attribute is true with regard to the second response is obtained, wherein the first unapproved attribute is different from the second unapproved attribute.

6. The apparatus according to claim 5, wherein an alias of the attribute value and an abbreviation of the attribute value are added to the recognition target words to increase a number of the recognition target words when the number of the recognition target words is less than or equal to a predetermined number.

7. A spoken dialogue method which retrieves desired data from a data storage unit storing data in association with attribute values concerning a plurality of attributes of the data, comprising:

performing speech recognition of an input speech of a user for a presented response by a spoken dialogue apparatus to obtain a recognition candidate in one attribute from recognition target words corresponding to the presented response, by a speech recognition unit;

updating a retrieval condition by use of a recognition candidate and retrieving data corresponding to the retrieval condition from the data storage unit to output as a retrieval result, by a retrieval unit;

calculating estimated numbers of data as costs concerning a first response and a second response based on the retrieval condition and the retrieval result, the data being included in retrieval result after narrowed retrieval performed in accordance with an input speech corresponding to the first response asking the user to input an attribute value of an attribute which has not been input and corresponding to the second response ascertaining the truth of an attribute value of an attribute which has been input, by a calculation unit; and

selecting a response having the lowest cost and presenting the response to the user, by a selection unit,

wherein the retrieval unit (a) updates the retrieval condition by adding the recognition candidate to the retrieval condition when the presented response corresponds to the first response, (b) updates the retrieval condition by maintaining an approved attribute value in the retrieval condition and deleting other attribute value of the attribute related to the presented response from the retrieval condition when the presented response corresponds to the second response and the recognition candidate indicates approval, and (c) updates the retrieval condition by deleting a denied attribute value form the retrieval condition and maintaining other attribute value of the attribute related to the presented response in the retrieval condition when the presented response corresponds to the second response and the recognition candidate indicates denial.

8. The method according to claim 7, wherein the second response ascertains the truth of an attribute value having a highest score provided by the speech recognition unit regarding the attribute related to the presented response.

9. The method according to claim 7, wherein the calculation unit further calculates a cost of a third response ascertaining the truth of an attribute value of the attribute which has not been input.

10. The method according to claim 7, wherein the retrieval unit modifies the retrieval condition by deleting, from the retrieval condition, all the attribute values but an attribute value previously ascertained to be true with regard to the second response when the retrieval result is null.

11. The method according to claim 7, wherein the speech recognition unit again performs speech recognition of a previous input speech in order to obtain a recognition candidate concerning a first unapproved attribute from the recognition target words when an input speech indicating that an attribute value of a second unapproved attribute is true with regard to the second response is obtained, wherein the first unapproved attribute is different from the second unapproved attribute.

12. The method according to claim 11, wherein an alias of the attribute value and an abbreviation of the attribute value are added to the recognition target words to increase a number of the recognition target words when the number of the recognition target words is less than or equal to a predetermined number.