Automated voice and speech labeling

• US 9,129,605 B2
• Filed: 03/14/2013
• Issued: 09/08/2015
• Est. Priority Date: 03/30/2012
• Status: Active Grant

• Current Assignee: SRC, Inc.
• Original Assignee: SRC, Inc.
• Inventors: Eller, David Donald, Morphet, Steven Brian, Boyett, Watson Brent
• Primary Examiner(s): Chawan, Vijay B
• Application Number:
  US13/828,856
• Publication Number:
  US 20130262111A1
• Time in Patent Office: 908 Days
• Field of Search: 704/235, 704/256.7, 704/260, 704/243, 704/231, 704/251, 704/257, 704/258, 704/240, 379/265.02, 379/309, 379/52, 379/88.16
• US Class Current: 1/1
• CPC Class Codes: G06F 16/685 using automatically derived transcript of audio data, e.g. lyrics speech recognition G10L15/00
• CPC Class Codes: G10L 15/10 using distance or distortion measures between unknown speech and reference templates
• 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

Abstract

A system and method for voice and speech analysis which correlates a speaker signal source and a normalized signal comprising measurements of input acoustic data to a database of language, dialect, accent, and/or speaker attributes in order to create a transcription of the input acoustic data.

Claims

1. A method for converting speech to text, comprising the steps of:

receiving a digital signal comprising a recorded spoken input;

obtaining at least one measurement of said digital signal, the measurement comprising a first measured portion of said recorded spoken input and a second measured portion of said recorded spoken input;

identifying at least one characteristic of said digital signal by comparing said first measured portion of said recorded spoken input to a first database of digital audio signal characteristics;

transcribing said first measured portion of said recorded spoken input using said at least one characteristic of said digital signal to create an initial transcription;

backfilling said first database of digital audio signal characteristics with at least one characteristic from a second database of digital audio signal characteristics;

identifying a second characteristic of said digital signal by comparing said second measured portion of said digital signal to said backfilled first database of digital audio signal characteristics;

transcribing said second measured portion of said recorded spoken input using said second characteristic of said digital signal.

2. The method of claim 1, further comprising the step of processing the received digital signal.

3. The method of claim 2, wherein the step of processing the received digital signal comprises the step of obtaining at least one measurement of said spoken input.

4. The method of claim 2, wherein the step of processing the received digital signal comprises the step of labeling vocal and non-vocal portions of said spoken input.

5. The method of claim 1, further comprising the step of normalizing said digital signal.

6. The method of claim 1, wherein said at least one characteristic of said digital signal is a characteristic of the speaker of said spoken input.

7. The method of claim 1, wherein at least one characteristic of said digital audio signal is used to produce said first level transcription.

8. The method of claim 1, further comprising the step of displaying said second transcription on a monitor.

19. The method of claim 1, wherein said second database of digital audio signal characteristics is populated with characteristics derived from a plurality of speakers.

9. A system for converting speech to text, the system comprising:

a digital audio signal comprising an encoding of a recorded spoken input;

means for obtaining at least one measurement of said digital audio signal, the measurement comprising a first measured portion of said recorded spoken input and a second measured portion of said recorded spoken input;

means for comparing said first measured portion of said recorded spoken input to a first database of digital audio signal characteristics;

means for identifying at least one characteristic of said digital audio signal based on said comparison;

means for transcribing said first measured portion of said spoken input using said at least one characteristic of the digital audio signal to create an initial transcription;

means for backfilling said first database of digital audio signal characteristics with at least one characteristic from a second database of digital audio signal characteristics;

means for identifying a second characteristic of said digital signal by comparing said second measured portion of said digital signal to said backfilled first database of digital audio signal characteristics; and

means for transcribing said second measured portion of said recorded spoken input using said second characteristic of said digital signal.

10. A system according to claim 9, further comprising means for filtering the digital audio signal.

11. A system according to claim 9, further comprising normalization means for normalizing the digital audio signal.

12. A system according to claim 9, comprising normalization means for normalizing prosodic speech features.

13. A system according to claim 9, comprising means for labeling the vocal and nonvocal portions of said digital audio signal.

14. A system according to claim 9, further comprising means identifying at least one characteristic of said digital audio signal.

15. A system according to claim 9, further comprising means for comparing characteristics to produce said initial transcript.

16. A system according to claim 9, further comprising means for constructing a multi-speaker feature map.

17. A system according to claim 9, further comprising means for using a multi-speaker feature map to backfill a single-speaker feature map.

18. A system according to claim 9, further comprising displaying means for displaying the second transcription on a monitor.

20. The system of claim 9, wherein said second database of digital audio signal characteristics is populated with characteristics derived from a plurality of speakers.