Transaction system based on a bidirectional speech channel by status graph building and problem detection for a human user

• US 5,860,059 A
• Filed: 03/05/1997
• Issued: 01/12/1999
• Est. Priority Date: 03/05/1996
• Status: Expired due to Term

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: US Philips Corporation (Koninklijke Philips N.V.)
• Inventors: Marti Del Olmo, Erique, Mehlan, Ralf, Aust, Harald, Marschner, Jens F., Schroer, Olaf, Brode, Holger W.
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): SAX, ROBERT L
• Application Number:
  US08/810,778
• Time in Patent Office: 678 Days
• Field of Search: 704/231, 704/275
• US Class Current: 704/231
• CPC Class Codes: G06F 3/16 Sound input; Sound output speech processing G10L
• CPC Class Codes: G10L 15/1815 Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 2015/221 Announcement of recognition results
• CPC Class Codes: H04M 2201/60 Medium conversion
• CPC Class Codes: H04M 3/493 Interactive information services, e.g. directory enquiries ; Arrangements therefor, e.g. interactive voice response [IVR] systems or voice portals

Abstract

A transaction system has machine recognition of speech. It has dialogue control fed by the recognition, and speech generation fed by the dialogue control for outputting question and verifier statements from a repertoire set. A human-machine dialogue is executed until the dialogue control has recognized a viable transaction formulation with a plurality of user-provided slot fillers to specify the transaction. Dialogue control builds a directed and loopless status graph with nodes that each have their own slot filler and associated metric, and are interrelated through logic relations. The building can amend a node'"'"'s metric and under control of conflict detection or lowering of a particular node'"'"'s metric, discard the node in question and its filler, including of derived nodes and also of one-to-one derival nodes of the discarded node. A subset of problems is detected for a particular item outputted by the speech generation for ascertaining a particular slot filler, as combined with a subsequent unsatisfactory recognition. Then the speech generation outputs a further question statement for ascertaining the latter slot filler.

Claims

1. A transaction system based on machine-based speech recognition, comprising a speech input, speech recognition means fed by said speech input, dialogue control means fed by said speech recognition means, and speech generation means fed by said dialogue control means for outputting question statements and verifier statements from a statement repertoire set, and for executing a human-machine dialogue with said system, until said dialogue control means has recognized a viable transaction formulation comprising a plurality of user-provided slot fillers for specifying said transaction,

characterized in that said dialogue control means comprise with respect to a particular transaction:

status graph building means for building one or more directed and loopless status graphs with nodes that each comprise a respective one of said slot fillers and an associated verifier metric, and which nodes are interrelated through logic combinations among their respective slot fillers, and said graph building means including amending means for amending a node'"'"'s metric, and discard means for under control of conflict detection or lowering of a particular node'"'"'s metric, discarding the node in question, including of all nodes derived from the so discarded node, and also one-to-one derival nodes of the discarded node,

problem detector means for detecting a particular subset of problems associated with a particular statement outputted by said speech generation means regarding a particular slot filler item, as combined with an associated subsequent human speech unsatisfactory recognition result, and thereupon controlling said speech generation means for outputting a subsequent question statement for ascertaining the latter slot filler.

2. A system as claimed in claim 1, wherein said subsequent question statement is of simplified structure as compared to said particular predecessor statement.

3. A transaction system as claimed in claim 1, wherein said discarding pertains also to all identical many-to-one derival nodes but excludes non-identical many-to-one derival nodes of any discarded node.

4. A transaction system as claimed in claim 1, wherein each recognized slot filler item is verified once by a directly subsequent verifier statement.

5. A transaction system as claimed in claim 1, wherein said problem detector is reactive to a score which is accumulated from a least a subset of the following events:

TOO_-- LOUD, TOO_-- QUIET, NOTHING_-- RECORDED, TOO_-- LONG, OVERLAP, NOTHING_-- UNDERSTOOD.

6. A transaction system as claimed in claim 1, for use as a trains timetable information system.

7. A transaction system as claimed in claim 1, for use as a telephone directory information system.

8. A method for implementing a human-machine transaction based on machine recognizing of speech, said method comprising the steps of:

receiving and recognizing human speech, controlling a dialog on the basis of said recognized speech, outputting in the form of machine speech various question statements and verifier statements from a statement repertoire set, and executing a human-machine dialog until recognizing a viable transaction formulation comprising a plurality of user-provided slot fillers for specifying said transaction,

being characterized in with respect to a particular transaction, comprising the steps of;

building one or more directed and loopless status graphs with nodes that each comprise a respective one of said slot fillers and an associated verifier metric, whilst interrelating said nodes by logic combinations among their respective slot fillers,

if applicable, amending said status graph as regarding a node'"'"'s metric and selectively, under control of conflict detection or lowering of a particular node'"'"'s metric, discarding the node in question, including of all nodes derived from the discarded node, and also one-to-one derival nodes of any discarded node,

detecting a particular subset of problems associated with a particular said statement regarding a particular slot filler, as combined with an associated subsequent unsatisfactory human speech recognition result, and thereupon outputting a subsequent question statement for ascertaining the latter slot filler.

9. A method as claimed in claim 8 wherein said subsequent question statement is of simplified structure as compared to said particular predecessor statement.

10. A method as claimed in claim 8, wherein said discarding pertains also to all identical many-to-one derival nodes but excludes non-identical many-to-one derival nodes of any discarded node.

11. A method as claimed in claim 8, wherein each recognized slot filler item is verified once by a directly subsequent verifier statement.

12. A method as claimed in claim 8, wherein said problem detecting is reactive to a score which is accumulated from a least a subset of the following events:

TOO_-- LOUD, TOO_-- QUIET, NOTHING_-- RECORDED, TOO_-- LONG, OVERLAP, NOTHING_-- UNDERSTOOD.

13. A method as claimed in claim 8, for use as a trains timetable information system.

14. A method as claimed in claim 8, for use as a telephone directory information system.