Speech recognition program mapping tool to align an audio file to verbatim text

• US 20030004724A1
• Filed: 04/05/2002
• Published: 01/02/2003
• Est. Priority Date: 02/05/1999
• Status: Abandoned Application

• Current Assignee: Custom Speech USA, Inc.
• Original Assignee: Custom Speech USA, Inc.
• Inventors: Scalpone, Stephen J., Kahn, Jonathan, Huttinger, Michael C.
• Application Number:
  US10/117,480
• Publication Number:
  US 20030004724A1
• Time in Patent Office: Days
• US Class Current: 704/260
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/065 Adaptation

Abstract

The invention includes a method to determine time location of at least one audio segment in an original audio file comprising: (a) receiving the original audio file; (b) transcribing a current audio segment from the original audio file using speech recognition software; (c) extracting a transcribed element and a binary audio stream corresponding to the transcribed element from the speech recognition software; (d) saving an association between the transcribed element and the corresponding binary audio stream; (e) repeating (b) through (d) for each audio segment in the original audio file; (f) for each transcribed element, searching for the associated binary audio stream in the original audio file, while tracking an end time location of that search within the original audio file; and (g) inserting the end time location for each binary audio stream into the transcribed element-corresponding binary audio stream association.

Claims

1. A method to determine time location of at least one audio segment in an original audio file comprising:

(a) receiving the original audio file;

(b) transcribing a current audio segment from the original audio file using speech recognition software;

(c) extracting a transcribed element and a binary audio stream corresponding to the transcribed element from the speech recognition software;

(d) saving an association between the transcribed element and the corresponding binary audio stream;

(e) repeating (b) through (d) for each audio segment in the original audio file;

(f) for each transcribed element, searching for the associated binary audio stream in the original audio file, while tracking an end time location of that search within the original audio file; and

(g) inserting the end time location for each binary audio stream into the transcribed element-corresponding binary audio stream association.

2. The method of claim 1 wherein searching includes removing any DC offset from the corresponding binary audio stream.

3. The method of claim 2, wherein removing any DC offset includes taking a derivative of the corresponding binary audio stream to produce a derivative binary audio stream.

4. The method of claim 3 wherein searching includes

taking a derivative of a segment of the original audio file to produce a derivative audio segment;

and

searching for the derivative binary audio stream in the derivative audio segment.

5. The method of claim 1 further including saving each transcribed element-corresponding binary audio stream association in a single file.

6. The method of claim 5 where the single file includes, for each word saved, a text for the transcribed element and a pointer to the binary audio stream.

7. The method of claim 5 wherein extracting is performed by using the Microsoft Speech API as an interface to the speech recognition software, wherein the speech recognition software does not return a word with a corresponding audio stream.

14. The method of claim 5 wherein the means for extracting is performed by using the Microsoft Speech API as an interface to the speech recognition software, wherein the speech recognition software does not return a word with a corresponding audio stream.

8. A system for determining a time location of at least one audio segment in an original audio file comprising:

means for receiving the original audio file;

means for transcribing a current audio segment from the original audio file using speech recognition software;

means for extracting a transcribed element and a binary audio stream corresponding to the transcribed element from the speech recognition software;

means for saving an association between the transcribed element and the corresponding binary audio stream;

means for searching for the associated binary audio stream in the original audio file, while tracking an end time location of that search within the original audio file; and

means for inserting the end time location for the binary audio stream into the transcribed element-corresponding binary audio stream association.

9. The method of claim 8 wherein the means for searching include means for removing any DC offset from the corresponding binary audio stream.

10. The method of claim 9, wherein the means for removing any DC offset include means for taking a derivative of the corresponding binary audio stream to produce a derivative binary audio stream.

11. The method of claim 10 wherein means for searching include means for taking a derivative of a segment of the original audio file to produce a derivative audio segment;

and means for searching for the derivative binary audio stream in the derivative audio segment.

12. The method of claim 8 further including means for saving each word-corresponding binary audio stream association in a single file.

13. The method of claim 12 where the single file includes, for each word saved, a text for the word and a pointer to the binary audio stream.

16. The system of claim 15 wherein the program uses a Microsoft Speech API.

15. A system for determining a time location of at least one audio segment in an original audio file comprising:

a storage device for storing the original audio file;

a speech recognition engine to transcribe a current audio segment from the original audio file;

a program that extracts a transcribed element and a binary audio stream corresponding to the transcribed element from the speech recognition software;

saves an association between the transcribed element and the corresponding binary audio stream into a session file;

searches for the binary audio stream audio stream in the original audio file; and

inserts the end time location for each binary audio stream into the transcribed element-corresponding binary audio stream association.