Speech recognition of character sequences

• US 8,700,397 B2
• Filed: 07/30/2012
• Issued: 04/15/2014
• Est. Priority Date: 10/30/2006
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Inventors: White, Kenneth D.
• Primary Examiner(s): Lerner, Martin
• Application Number:
  US13/562,114
• Publication Number:
  US 20120296653A1
• Time in Patent Office: 624 Days
• Field of Search: 704/235, 704/236, 704/239, 704/240, 704/251, 704/255, 704/254
• US Class Current: 704/236
• CPC Class Codes: G10L 15/08 Speech classification or search

Abstract

A method of and a system for processing speech. A spoken utterance of a plurality of characters can be received. A plurality of known character sequences that potentially correspond to the spoken utterance can be selected. Each selected known character sequence can be scored based on, at least in part, a weighting of individual characters that comprise the known character sequence.

Claims

1. A method comprising:

receiving a spoken utterance of a plurality of uttered characters;

determining an identified character sequence by determining corresponding identified characters for individual ones of the plurality of uttered characters;

selecting a plurality of known character sequences that potentially correspond to the identified character sequence; and

for each selected known character sequence, scoring such known character sequence, using a processor, based at least in part on a weighting of individual characters that comprise the known character sequence, wherein scoring the known character sequence comprises;

determining whether a selected character of the known character sequence matches a selected character of the identified character sequence;

when the selected character of the known character sequence matches the selected character of the identified character sequence;

selecting a value that corresponds to the selected character of the known character sequence; and

adding the selected value to a cumulative score associated with the known character sequence; and

when the selected character of the known character sequence does not match the selected character of the identified character sequence;

making no changes to the cumulative score associated with the known character sequence based upon the selected character of the known character sequence.

2. The method of claim 1, wherein the weighting is based at least in part on frequencies of utterance of the individual characters.

3. The method of claim 1, wherein the selected value is pre-assigned to the character in a character set.

4. The method of claim 1, further comprising:

selecting the known character sequence having the highest cumulative score; and

communicating the selected known character sequence having the highest cumulative score to an application that receives recognized speech.

5. The method of claim 1, wherein selecting the plurality of known character sequences comprises selecting character sequences from a limited set of character sequences.

6. A method comprising:

receiving a spoken utterance of a plurality of uttered characters;

determining an identified character sequence by determining corresponding identified characters for individual ones of the plurality of uttered characters;

selecting a plurality of known character sequences that potentially correspond to the identified character sequence; and

for each selected known character sequence, scoring such known character sequence, using a processor, based at least in part on a weighting of individual characters that comprise the known character sequence, wherein scoring the known character sequence comprises;

for each individual character of the known character sequence, weighting the individual character based on a unigram frequency with which the individual character is known to be uttered, wherein said weighting comprises;

determining whether a selected character of the known character sequence matches a selected character of the identified character sequence; and

when the selected character of the known character sequence matches the selected character of the identified character sequence;

selecting a value that corresponds to the selected character of the known character sequence; and

adding the selected value to a cumulative score associated with the known character sequence; and

when the selected character of the known character sequence does not match the selected character of the identified character sequence;

making no changes to the cumulative score associated with the known character sequence based upon the selected character of the known character sequence.

7. The method of claim 6, wherein the selected value is pre-assigned to the character in a character set.

8. The method of claim 6, further comprising:

selecting the known character sequence having the highest cumulative score; and

communicating the selected known character sequence having the highest cumulative score to an application that receives recognized speech.

9. The method of claim 6, wherein selecting the plurality of known character sequences comprises selecting character sequences from a limited set of character sequences.

10. A non-transitory computer-readable storage medium having computer-usable program code that, when executed, performs a method comprising:

receiving a spoken utterance of a plurality of uttered characters;

determining an identified character sequence by determining corresponding identified characters for individual ones of the plurality of uttered characters;

selecting a plurality of known character sequences that potentially correspond to the identified character sequence; and

for each selected known character sequence, scoring such known character sequence, based at least in part on a weighting of individual characters that comprise the known character sequence, wherein scoring the known character sequence comprises;

determining whether a selected character of the known character sequence matches a selected character of the identified character sequence;

when the selected character of the known character sequence matches the selected character of the identified character sequence;

selecting a value that corresponds to the selected character of the known character sequence; and

adding the selected value to a cumulative score associated with the known character sequence; and

when the selected character of the known character sequence does not match the selected character of the identified character sequence;

making no changes to the cumulative score associated with the known character sequence based upon the selected character of the known character sequence.

11. The non-transitory computer-readable storage medium of claim 10, wherein the weighting is based at least in part on frequencies of utterance of the individual characters.

12. The non-transitory computer-readable storage medium of claim 11, wherein the frequencies of utterance are unigram frequencies.

13. The method of claim 10, wherein the selected value is pre-assigned to the character in a character set.

14. The non-transitory computer-readable storage medium of claim 10, wherein the method further comprises:

selecting the known character sequence having the highest cumulative score; and

communicating the selected known character sequence having the highest cumulative score to an application that receives recognized speech.

15. The non-transitory computer-readable storage medium of claim 10, wherein selecting the plurality of known character sequences comprises selecting character sequences from a limited set of character sequences.