Robust speech recognition with data bank accession organized by semantic attribute

• US 7,587,322 B2
• Filed: 10/31/2001
• Issued: 09/08/2009
• Est. Priority Date: 11/03/2000
• Status: Expired due to Fees

• Current Assignee: Voicecom Innovations GmbH
• Original Assignee: Voicecom Innovations GmbH
• Inventors: Plankensteiner, Peter, Harbeck, Stefan, Schimmer, Klaus
• Primary Examiner(s): Armstrong; Angela A
• Application Number:
  US10/415,709
• Publication Number:
  US 20040148167A1
• Time in Patent Office: 2,869 Days
• Field of Search: 704/9, 704/231, 704/252, 704/257, 704/275
• US Class Current: 704/257
• CPC Class Codes: G10L 15/1822 Parsing for meaning understanding
• CPC Class Codes: G10L 2015/223 Execution procedure of a spoken command
• CPC Class Codes: G10L 2015/228 of application context

Abstract

A method for controlling an information system during the output of stored information segments via a signaling device (50a). Useful information is stored in a database (32) for being requested, from which information at least one information segment is specified as a first data segment (W1) via a first voice signal (s_a(t),s_a(z)) and is provided via a control output (20,40,50;50a) or is converted (50b) into a control signal for a technical device (G). The information is organized in the database such that an initially limited first information area (32a) of stored information is accessible (4,4a,4b) to said voice signal, for selecting the specified information segment therefrom. A further information area (32b,32c,32d) of said database (32) is activated (59,70,4c,4d) as a second information area, if the information segment (W1) corresponding to a first voice signal segment (s1) of said first voice signal (s_a(t) is not contained in said first information area (32a). When accessing information of the database, a robust word recognition is obtained and the request is successfully processed within a short time.

Claims

1. A method for controlling an information system during an output of stored information segments via a signaling device, the method comprising the steps of

storing information in a database for being requested;

providing a first voice signal having a first voice signal segment and providing a voice recognition module for decoding the first voice signal;

organizing the information in said database such that a limited first information area of stored information is initially accessible to said first voice signal segment;

specifying at least one information segment from the stored information as a first data segment by a decoded first voice signal segment and provide the specified information segment as output;

activating a further information area of said database as a second information area from the stored information, when said first data segment corresponding to the first voice signal segment of said first voice signal is not contained in said limited first information area, wherein the activation of said second information area of the database is at least co-dependent on a semantic category of said voice signal segment;

said semantic category being determined by the voice recognition module when decoding the voice signal segment of the first voice signal, to associate the semantic category to the voice signal segment.

2. The method of claim 1, wherein the output of stored information is communicated as a control output.

3. The method of claim 1, wherein said output of information is controlling a technical device.

4. The method of claim 1, wherein after activating said further information area, a third information area is activated if said specified information segment is also not available from said further activated information area.

5. The method of claim 1, wherein said first information area or said second information area is a real subset of the information stored in said database.

6. The method of claim 1, wherein said first limited information area of stored information of said database is at least partly determined according to a user profile.

7. The method of claim 1, wherein there is provided a threshold value or a confidence value calculated from said voice signal by a decoder, to allow a comparison.

8. The method of claim 1, wherein a status of not-containing said first data segment results from a recognition of acoustic models, each of said acoustic models being associated to an individual semantic category.

9. The method of claim 1, wherein said voice signal is decoded in a decoder using acoustic models, for providing a confidence value in said semantic category for said first voice signal segment of the first voice signal.

10. The method of claim 1, wherein a threshold value is provided for testing a confidence value determined by a voice recognition as an acoustic decoder, wherein said confidence value is compared to said threshold value.

11. The method of claim 10, wherein said comparison with said threshold value is effected, and depending on the comparison, the semantic category of a certain word being related to said first voice signal segment is evaluated.

12. The method of claim 1, wherein an information segment specified by said first voice signal segment is output via a control output, when a voice segment characterizing said specified information segment is found as a word in one of said first limited information area, said second information area and another information area activated later.

13. The method of claim 12, wherein the activation of said second information area depends on the semantic category of said voice signal segment, which voice signal segment having no correspondent in said limited first information area.

14. The method of claim 1, wherein the activation of said further information area depends on the semantic category of said voice signal segment, which voice signal segment had no corresponding word in said limited first information area as the voice signal segment could not be determined with a sufficient level of confidence.

15. The method of claim 1, wherein after activating said further information area, only a shortened voice signal segment is evaluated in a decoder, for specifying in said further information area an information segment corresponding to said voice signal segment.

16. The method of claim 15, wherein the shortened voice signal segment is the voice signal segment of said first voice signal, evaluated in the decoder, for specifying the corresponding information segment.

17. The method of claim 1, wherein after detecting a confidence value and a certain semantic category of said voice signal segment, a perceivable signal is output, said perceivable signal being dependent on the semantic category, if the confidence value is too small.

18. The method of claim 1, wherein said activated further information area is different from said first limited information area by one of larger, offset, and fully different.

19. The method of claim 1, wherein after evaluating the semantic category by said voice recognition module, a further voice signal segment is supplied for being evaluated having less perplexity.

20. The method of claim 19, wherein further voice signal segment supplied for being evaluated, has temporal length thereof being shortened with respect to said first voice signal evaluated by said voice recognition module.

21. The method of claim 1, wherein the activation of said second information area depends on the semantic category of said voice signal segment, which voice signal segment could not be determined, recognized, specified or selected with sufficient confidence.

22. A method for controlling an information system during an output of stored information segments via a signaling device, the method comprising the steps of

storing information in a database suitable for being requested;

providing a voice recognition system;

specifying an information segment as a first data segment via a first voice signal provided to said voice recognition system from a user;

providing the first data segment as a control output;

organizing the information in said database such that initially only a first information area of stored information is accessible to said first voice signal, for selecting said specified information segment from said first information area;

associating a semantic category as one of a group of properties to a voice signal segment within said voice signal with no dialogue requests to said user, when an information segment corresponding to said voice signal segment is not contained in said first information area; and

activating a second information area,

wherein said second activated information area is different from said first accessible information area and dependent on the semantic category, as associated to said voice signal segment.

23. The method of claim 22, wherein after evaluating the semantic category by said voice recognition system, a further voice signal segment is supplied for being evaluated1 a temporal length of the further segment being shortened with respect to said first voice signal evaluated by said voice recognition system.

24. A method for controlling a technical device via a voice initiated action, comprising the steps of

storing information in a database suitable for being requested;

providing a voice recognition system;

specifying an information segment as a first data segment via a first voice signal provided to said voice recognition system from a user;

providing the first data segment as a converted control signal for said technical device;

organizing the information in said database such that initially only a first information area of stored information is accessible to said first voice signal, for selecting said specified information segment from said first information area;

associating a semantic category as one of a group of properties to a voice signal segment within said voice signal with no dialogue requests to said user that provided the first voice signal, when an information segment corresponding to said voice signal segment of said first voice signal is not contained in said first information area; and

activating a second information area, wherein said second activated information area is different from said first information area, and dependent on the semantic category, as associated to said voice signal segment.

25. A method for controlling a technical device via a voice initiated action, the method comprising the steps of

storing information in a database for being requested;

providing a first voice signal having a first voice signal segment and providing a voice recognition module for decoding the voice signal;

organizing the information in said database such that a limited first information area of stored information is initially accessible to said first voice signal segment;

specifying at least one information segment from the stored information as a first data segment by a decoded first voice signal segment and provide the specified information segment as output;

activating a further information area of said database as a second information area from the stored information, when said first data segment corresponding to the first voice signal segment of said first voice signal is not contained in said limited first information area, wherein the activation of the second information area of the database is at least co-dependent on a semantic category of said voice signal segment;

said semantic category being determined by the voice recognition module when decoding the voice signal segment of the first voice signal, to associate the semantic category to the voice signal segment.

26. The method of claim 25, wherein the output is converted into a control signal for said technical device.