System for speech recognition and correction, correction device and method for creating a lexicon of alternatives

• US 8,447,602 B2
• Filed: 03/22/2004
• Issued: 05/21/2013
• Est. Priority Date: 03/26/2003
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Nuance Communications Austria Gmbh (Microsoft Corporation)
• Inventors: Bartosik, Heinrich Franz, Meyer, Carsten
• Primary Examiner(s): PULLIAS, JESSE SCOTT
• Application Number:
  US10/550,967
• Publication Number:
  US 20070033026A1
• Time in Patent Office: 3,347 Days
• Field of Search: 704/10, 704/257, 704/E15.018, 704/E15.044, 704/235, 704/231
• US Class Current: 704/235
• CPC Class Codes: G06F 40/232 Orthographic correction, e.g. spell checking or vowelisation
• CPC Class Codes: G06F 40/242 Dictionaries
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 2015/221 Announcement of recognition results

Abstract

In a speech recognition and correction system which comprises at least one speech recognition device (1) to which a spoken text (GT) can be fed, it being possible for said spoken text to be transcribe into a recognized text (ET), and a correction device (3) for correcting the text (ET) recognized by the at least one speech recognition device (1), said correction device being connected to the at least one speech recognition device (1) via a data network (2) for the transmission of the recognized text (ET) and where appropriate of the spoken text (GT), the correction device (3) has a lexicon of alternatives (23) which contains word parts, words and word sequences that can be displayed (22) by the correction device (3) as alternatives to individual word parts, words and word sequences of the recognized text.

Claims

1. A speech recognition and correction system comprising:

at least one speech recognition device configured to transcribe a spoken text into a recognized text; and

a correction device configured to;

correct the recognized text, said correction device being connected to the at least one speech recognition device via a data communications medium for the transmission of the recognized text and/or of the spoken text;

access a lexicon of alternatives, the lexicon of alternatives comprising a plurality of entries, wherein the plurality of entries include one or more alternatives determined from one or more sources of knowledge that are independent of an analysis by an acoustic model and a language model used by the at least one speech recognition device during transcription of the spoken text, wherein the one or more sources of knowledge include confusion information compiled from corrected texts and associated recognized texts; and

display at least some of the plurality of entries as a list of alternatives to individual word parts, words and/or word sequences of the recognized text.

2. A speech recognition and correction system as claimed in claim 1, wherein the correction device is further configured to:

update the list of alternatives for at least some of the plurality of entries in the lexicon of alternatives displayed for a particular individual word part, word, and/or word sequence based, at least in part, on a number of times that the correction device previously corrected the particular individual word part, word, and/or word sequence with a text element replacement selected by a user, wherein the list of alternatives is updated only when the number of times is at least two times.

3. A correction device for correcting a text recognized by a speech recognition device, the correction device comprising:

a storage device configured to store a lexicon of alternatives comprising a plurality of entries, wherein the plurality of entries include one or more alternatives determined from one or more sources of knowledge that are independent of an analysis by an acoustic model and a language model used by the speech recognition device, wherein the one or more sources of knowledge include confusion information compiled from corrected texts and associated recognized texts; and

at least one processor configured to;

display at least some of the plurality of entries as a list of alternatives to individual word parts, words and/or word sequences of the recognized text.

4. A correction device as claimed in claim 3, further comprising:

an analyzer configured to analyze selected text passages of the recognized text by using character chain comparison or syntactic analysis, and to determine alternatives to the selected text passages from the lexicon of alternatives.

5. A correction device as claimed in claim 4, wherein the analyzer can be activated by a user of the correction device.

6. A correction device as claimed in claim 4, wherein the analyzer determines selected text passages from a cursor position or a marking information of a text processing program.

7. A correction device as claimed in claim 4, wherein the analyzer determines selected text passages from a time position of the spoken text and its association with the recognized text.

8. A correction device as claimed in claim 3, wherein the at least one processor is further configured to:

update the list of alternatives for at least some of the plurality of entries in the lexicon of alternatives displayed for a particular individual word part, word, and/or word sequence based, at least in part, on information about at least one previous correction made by the correction device for the particular individual word part, word, and/or word sequence with a text element replacement selected by a user, wherein the list of alternatives is updated only when at least a predetermined degree of phonetic similarity exists between the particular individual word part, word, and/or word sequence and a text replacement in the at least one previous correction.

9. The correction device of claim 3, wherein the one or more alternatives are further determined based, at least in part, on confusion information for a particular speaker of the associated recognized texts.

10. The correction device of claim 3, wherein the one or more alternatives are further determined based, at least in part, on syntactic information.

11. The correction device of claim 3, wherein the one or more alternatives are further determined based, at least in part, on phonetic sequence information.

12. The correction device of claim 3, wherein the confusion information comprises confusion statistics.

13. A computer-implemented method of creating an entry in a lexicon of alternatives used to correct recognized text transcribed from a spoken text by a speech recognition device, the method comprising:

examining, by at least one processor, at least one source of knowledge that is independent of an acoustic model and a language model used by the speech recognition device, wherein the at least one source of knowledge includes confusion information compiled from corrected texts and associated recognized texts, wherein the at least source of knowledge is examined with respect to text elements, including word parts, words and/or word sequences associated with the confusion information contained therein to identify text elements that can be confused with one another; and

including the text elements that can be confused with one another as a list of alternatives in the entry of the lexicon of alternatives.

14. A computer-implemented method as claimed in claim 13, further comprising:

determining the text element replacements made in a corrected text with respect to the recognized text transcribed by the speech recognition device; and

recording the text element replacements as alternatives in the lexicon of alternatives.

15. A computer-implemented method as claimed in claim 14, further comprising:

evaluating a frequency of each text element replacement, and

recording the text element replacements as alternatives in the lexicon of alternatives only when a predetermined lower limit value of the frequency, expressed by an absolute number of the text element replacements or the ratio of number of the text element replacements with respect to the overall number of text elements examined or with respect to an overall occurrence of a given text element is exceeded.

16. A computer-implemented method as claimed in claim 15, further comprising:

evaluating a frequency of each text element examined in the at least one source of knowledge; and

recording the text element replacements as alternatives in the lexicon of alternatives only when a predetermined upper limit value of the frequency, expressed by an absolute number of the text element replacements or a ratio of a number of the text element replacements with respect to an overall number of text elements examined, is not reached.

17. A computer-implemented method as claimed in claim 14, further comprising:

analyzing the acoustic similarity of the text element replacements; and

recording the text element replacements as alternatives in the lexicon of alternatives only when the text element replacements have a predetermined degree of phonetic similarity.

18. A computer-implemented method as claimed in claim 14, further comprising:

analyzing time positions of the text element replacements with respect to the spoken text; and

recording the text element replacements as alternatives in the lexicon of alternatives only when there is a corresponding text element in the spoken text that is similar in terms of time.

19. A computer-implemented method as claimed in claim 13, wherein the lexicon of alternatives comprises a plurality of entries, the method further comprising:

subdividing the plurality of entries according to speech.

20. A computer-implemented method as claimed in claim 13, wherein the lexicon of alternatives comprises a plurality of entries, the method further comprising:

subdividing the plurality of entries according to technical field or field of application.

21. A computer-implemented method as claimed in claim 13, wherein the lexicon of alternatives comprises a plurality of entries, the method further comprising:

subdividing the plurality of entries according to author of the spoken text or a corrected text.

22. A computer-implemented method as claimed in claim 13, wherein the lexicon of alternatives is adapted online during a correction of recognized texts.

23. A computer-implemented method as claimed in claim 13, wherein the at least one source of knowledge that is independent of an acoustic model and a language model used by the speech recognition device includes text files specific to a field of application.

24. A computer-implemented method as claimed in claim 13, further comprising:

updating the list of alternatives in the entry based, at least in part, on whether a frequency of previous corrections of the recognized text with text element replacements selected by a user is within predetermined bounds.