Speech recognition tuning tool

• US 9,183,834 B2
• Filed: 07/22/2009
• Issued: 11/10/2015
• Est. Priority Date: 07/22/2009
• Status: Active Grant

• Current Assignee: Cisco Technology, Inc. (Cisco Systems, Inc.)
• Original Assignee: Cisco Technology, Inc. (Cisco Systems, Inc.)
• Inventors: Gatzke, Alan D., Maas, Michael T., Bloom, Ryan L., Lindborg, Jeff B.
• Primary Examiner(s): ROBERTS, SHAUN A
• Application Number:
  US12/507,126
• Publication Number:
  US 20110022386A1
• Time in Patent Office: 2,302 Days
• Field of Search: 704/235, 704/E15.001, 704/E15.043, 704/E17.016
• US Class Current: 1/1
• CPC Class Codes: G10L 15/06 Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice G10L15/14 takes precedence
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 2015/0631 Creating reference templates; Clustering
• CPC Class Codes: H04M 2201/40 using speech recognition speech recognition per se G10L15/00
• 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
• CPC Class Codes: H04M 3/533 Voice mail systems

Abstract

Systems and methods for tuning a dictionary of a speech recognition system includes accessing a voice mail record of a user, accessing a recorded audio file of a name of the user in the voice mail record spoken by the user, providing the audio file to a speech recognition system, processing the audio file in the speech recognition system and obtaining a text result, determining whether a confidence score of the text result is below a predetermined threshold, and adding, at least, the name of the user to a list of low confidence names. Alternate spellings for the low confidence names can then be added to the dictionary.

Claims

1. A method, comprising:

accessing a voice mail record of a user within a voice mail system;

accessing a recorded audio file of a name of the user in the voice mail record spoken by the user;

providing the audio file to a speech recognition system that is operable with an automated attendant;

processing the audio file in the speech recognition system and obtaining a text result;

determining whether a confidence score of the text result is below a predetermined threshold;

adding, at least, the name of the user to a list of low confidence names when the confidence score is below the predetermined threshold;

when the name of the user is listed in the list of low confidence names, storing a plurality of actual alternate spellings for the name of the user, wherein the plurality of actual alternate spellings are accessible to the speech recognition system and are received via a user interface configured to be presented to an administrator of the automated attendant;

receiving a voice call at the automated attendant including receiving a voice command comprising a spoken name of the user; and

processing the spoken name of the user including comparing a spelled name result generated by the speech recognition system to the plurality of actual alternate spellings previously stored to identify the user.

2. The method of claim 1, further comprising obtaining a plurality of text results for each name in list of low confidence names.

3. The method of claim 2, further comprising obtaining an n-best list of low confidence names.

4. The method of claim 1, further comprising adding a link to the audio file in the list of low confidence names.

5. The method of claim 1, further comprising listing the confidence score along with the name of the user in the list of low confidence names.

6. The method of claim 1, further comprising repeating the method when a number of new users reaches a predetermined threshold.

7. The method of claim 1, further comprising receiving a request to play the audio file, the request having been initiated from the list of low confidence names.

8. The method of claim 1, wherein storing is initiated from an administrator'"'"'s user interface.

9. The method of claim 1, further comprising configuring the predetermined threshold.

10. The method of claim 1, further comprising repeating the method for each name in a directory of names.

11. An apparatus, comprising:

a speech recognition tuning tool configured to be in communication with a voice mail system, a low confidence audio recording database, and a speech to text converter, the speech recognition tuning tool having a controller,

wherein the controller is configured to;

access an audio file of a name of the user in the voice mail system;

provide the audio file to the speech to text converter operable with an automated attendant;

determine whether a confidence score of a text result from the speech to text converter is below a predetermined threshold;

add, at least, the name of the user to a list of low confidence names in the low confidence audio recording database when the confidence score is below the predetermined threshold;

receive, via a user interface configured to be presented, via a display, to an administrator of the automated attendant, and store a plurality of actual alternate spellings for the name of the user, wherein the plurality of actual alternate spellings are accessible to the controller;

receive a voice call at the automated attendant including receiving a voice command comprising a spoken name of the user; and

process the spoken name of the user including comparing a spelled name result generated by the speech to text converter to the plurality of actual alternate spellings previously stored to identify the user.

12. The apparatus of claim 11, wherein the controller is configured to save the alternate spellings in a dictionary.

13. The apparatus of claim 11, wherein the controller is configured to present to a user a link or path to the audio file.

14. The apparatus of claim 11, wherein the controller is configured to cause the audio file to be played.

15. Logic encoded in one or more non-transitory media for execution and when executed operable to:

access an audio file of a name of the user in a voice mail system;

provide the audio file to a speech to text converter operable with an automated attendant;

determine whether a confidence score of a text result from the speech to text converter is below a predetermined threshold;

add, at least, the name of the user to a list of low confidence names when the confidence score is below the predetermined threshold;

receive, via a user interface configured to be presented to an administrator of the automated attendant, and store a plurality of actual alternate spellings for the name of the user as a result of the name having been added to the list of low confidence names;

receive a voice call at the automated attendant including receiving a voice command comprising a spoken name of the user; and

process the spoken name of the user including comparing a spelled name result generated by the speech to text converter to the plurality of actual alternate spellings previously stored to identify the user.

16. The logic of claim 15, wherein the logic is further operable to save the alternate spellings in a dictionary.

17. The logic of claim 15, wherein the logic is further operable to present to a user a link or path to the audio file along with the name in the list of low confidence names.