Method and system for improving speech recognition accuracy by use of geographic information

• US 8,510,032 B2
• Filed: 08/06/2012
• Issued: 08/13/2013
• Est. Priority Date: 07/27/2009
• Status: Active Grant

• Current Assignee: Robert Bosch GmbH
• Original Assignee: Robert Bosch GmbH
• Inventors: Shen, Zhongnan, Weng, Fuliang, Feng, Zhe
• Primary Examiner(s): ARTHUR JEANGLAUDE, GERTRUDE
• Application Number:
  US13/567,242
• Publication Number:
  US 20120303267A1
• Time in Patent Office: 372 Days
• Field of Search: 701/400, 701/409, 701/418, 701/427, 701/431, 701/532, 701/488, 704/231, 704/E15.001, 704/251, 340/995.17
• US Class Current: 701/400
• CPC Class Codes: G10L 15/1815 Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning
• CPC Class Codes: G10L 2015/228 of application context

Abstract

A method for speech recognition includes providing a source of geographical information within a vehicle. The geographical information pertains to a current location of the vehicle, a planned travel route of the vehicle, a map displayed within the vehicle, and/or a gesture marked by a user on a map. Words spoken within the vehicle are recognized by use of a speech recognition module. The recognizing is dependent upon the geographical information.

Claims

1. A method for speech recognition, the method comprising the steps of:

providing a source of geographical information within a vehicle, the geographical information pertaining to at least one of;

a current location of the vehicle;

a planned travel route of the vehicle; and

a map displayed within the vehicle;

recognizing words spoken within the vehicle by use of a speech recognition module, the recognizing being dependent upon the geographical information; and

extracting a plurality of candidate interpretations from the geographical information, the recognizing being biased to a greater degree toward interpreting one of the spoken words to be a selected said candidate interpretation extracted from geographical information pertaining to the planned travel route of the vehicle than toward interpreting one of the spoken words to be a selected said candidate interpretation extracted from geographical information pertaining to one of the current location of the vehicle and a map displayed within the vehicle.

2. The method of claim 1 wherein the source of geographical information comprises a navigation system.

3. The method of claim 1 wherein at least one of the candidate interpretations is a name of an entity associated with said at least one of:

a current location of the vehicle;

a planned travel route of the vehicle; and

a map displayed within the vehicle.

4. The method of claim 1 wherein the geographical information pertains to a subset of the displayed map, the subset being manually marked by a user.

5. The method of claim 4 wherein the subset comprises an area marked by the user on a touch screen display, a boundary of the area being modeled as a closed loop, a point or a line.

6. A method for speech recognition, the method comprising the steps of:

automatically providing a first category of geographical information to a user;

deriving a second category of geographical information from inputs provided by the user at least one of orally and manually via a user interface;

extracting a plurality of candidate words from the geographical information, first ones of the candidate words being extracted from the first category of geographical information, and second ones of the candidate words being extracted from the second category of geographical information; and

recognizing words spoken within the vehicle by use of a speech recognition module, the words being spoken after the inputs have been provided by the user, the recognizing including selecting at least one word from the candidate words, the second candidate words being given more weight in the selecting than are the first candidate words.

7. The method of claim 6 wherein the first category of geographical information is automatically provided by a navigation system.

8. The method of claim 6 wherein the second category of geographical information is derived from at least one navigation command orally provided by the user.

9. The method of claim 6 wherein the second category of geographical information is derived from an area of a map displayed in the vehicle, the map area being manually marked by the user.

10. The method of claim 9 comprising the further steps of:

dividing the map into a plurality of sections; and

creating an index associating each of the sections with names of entities with the section, the extracting of the second candidate words including using the index to identify all of the entity names associated with each of the sections that are at least partially overlapped by the manually marked area of the map.

11. The method of claim 6 wherein the candidate words consist essentially of candidate entity names.

12. The method of claim 6 comprising the further step of using a speech recognition module to create a list of preliminary candidate words, the list of preliminary candidate words being created independent of the geographical information, the first candidate words being given more weight in the selecting than are the preliminary candidate words that are identified independent of the geographical information.

13. The method of claim 6 wherein the second category of geographical information is derived from a first input spoken by the user and a second input spoken by the user, the first input being spoken by the user before the second input is spoken by the user, a first one of the second candidate words being extracted from said second geographical information derived from the first spoken input provided by the user, a second one of the second candidate words being extracted from said second geographical information derived from the second spoken input provided by the user, the second one of the second candidate words being given greater weighting in the recognizing than is the first one of the second candidate words.

14. A method for speech recognition, the method comprising the steps of:

displaying within a vehicle a map of a first geographical area;

deriving geographical information from a second geographical area manually defined by a user on a touch screen on which the map is displayed, the second geographical area being a subset of the first geographical area; and

recognizing words spoken within the vehicle by use of a speech recognition module, the recognizing being dependent upon the geographical information derived from the second geographical area.

15. The method of claim 14 comprising the further step of deriving first geographical information from the first geographical area, the geographical information derived from the second geographical area comprising second geographical information, wherein at least one portion of the first geographical information is omitted from the second geographical information, the second geographical information being given greater weighting in the recognizing than is the first geographical information.

16. The method of claim 14 wherein the second geographical area is defined by a user touching a portion of the touch screen overlapping the map, the user touching the screen in a pattern modeled by a processor as a defining the second geographical area.

17. The method of claim 16 wherein the map is divided into a plurality of rectangular sections, the second geographical area comprising each said section that is at least partially overlapped by the pattern.

18. The method of claim 14 comprising the further steps of:

extracting names of entities within the second geographical area; and

presenting the extracted names to the user as candidate interpretations of the words spoken.

19. The method of claim 18 wherein the recognizing is biased toward interpreting the words spoken as at least one of the extracted names.

20. The method of claim 14 comprising the further step of extracting a predetermined number of names of entities within the second geographical area, wherein the second geographical information comprises the predetermined number of extracted names.