System and method for facilitating transcription of audio recordings, with auditing

• US 7,706,520 B1
• Filed: 11/08/2005
• Issued: 04/27/2010
• Est. Priority Date: 11/08/2005
• Status: Expired due to Fees

• Current Assignee: Serenova LLC (Lifesize Communications Incorporated)
• Original Assignee: Liveops Incorporated
• Inventors: Waterson, Christopher R., Tabb, Lloyd William, Toy, Michael
• Primary Examiner(s): Smits; Talivaldis Ivars
• Assistant Examiner(s): BORSETTI, GREG
• Application Number:
  US11/269,296
• Time in Patent Office: 1,631 Days
• Field of Search: 704/235, 704/236, 704/274, 379/265.06, 379/265.03, 379/265.11, 379/265.07, 705/11
• US Class Current: 379/265.06
• CPC Class Codes: H04M 2201/40 using speech recognition speech recognition per se G10L15/00
• CPC Class Codes: H04M 2201/60 Medium conversion
• CPC Class Codes: H04M 3/42221 Conversation recording systems at the subscriber's set H04M1/656

Abstract

A system and method for facilitating transcription of an audio recording, with auditing. An IVR (Interactive Voice Response) unit makes a set of audio recordings of a caller'"'"'s input during a call. The set of audio recordings is stored in a database and queued for transcription by remote agents. The recordings are forwarded to a first agent who listens to the audio and completes a form with text corresponding to the audio. The system then determines whether to audit the first agent'"'"'s transcription by having another agent transcribe the same set of recordings. The recordings may be re-queued and transcribed multiple times, until two (or more) of the transcriptions match, or until some threshold number of transcriptions have been obtained. An agent may contact the caller (e.g., via telephone or electronic mail) if a recording cannot be fully transcribed.

Claims

1. An automated computer-implemented method of facilitating and auditing the transcription of recorded audio, the method being executable as logic instructions on a computer device, comprising:

(a) queuing a set of audio recordings on a computer device for transcription by a first remote agent;

(b) receiving a first transcription of the set of audio recordings on said computer device from the first remote agent and determining whether to audit the first transcription by calculating an audit rate a based on a current error rate ε

of the first agent and a desired level of accuracy λ

;

(c) if audit rate is higher than a threshold rate, re-queuing the set of audio recordings on said computer device for transcription by a second remote agent among a plurality of remote agents; and

(d) receiving a next transcription on said computer device of the set of audio recordings from the second remote agent;

wherein audit rate α

indicates a frequency with which a transcript should be audited;

said current error rate ε

of the first agent represents a probability that the first transcription is incorrect; and

said desired level of accuracy λ

reflects a probability that a transcription is correct.

2. The method of claim 1, wherein (e) comprises repeating (c)-(d) until:

at least two transcriptions of the set of audio recordings match;

or

a threshold number of transcriptions of the set of audio recordings has been received.

3. The method of claim 1, further comprising, prior to (a):

receiving a call at an IVR (Interactive Voice Response) unit from a caller; and

generating the set of audio recordings to record input from the caller.

4. The method of claim 3, further comprising:

facilitating a communication connection between an agent and the caller to elicit information that cannot be retrieved from the set of audio recordings.

5. The method of claim 1, wherein:

said audit rate α

is calculated as α

≧

(λ

+ε

−

1)/(ε

−

ε

ⁿ); and

wherein n>

1 and represents a maximum number of transcriptions of the set of audio recordings to perform.

6. The method of claim 1, wherein said determining further comprises:

virtually flipping a weighted coin based on said audit rate α

.

7. The method of claim 1, further comprising:

serving the first remote agent a form for inputting the first transcription.

8. The method of claim 1, wherein (c) comprises re-queuing the set of audio recordings for transcription by a second remote agent without informing the first remote agent of the re-queuing.

9. A computer readable medium storing instructions that, when executed by a computer, cause the computer to perform a method of facilitating transcription of recorded audio, with auditing, the method comprising:

(a) queuing a set of audio recordings for transcription by a first remote agent;

(b) receiving a first transcription of the set of audio recordings from the first remote agent and determining whether to audit the first transcription by calculating an audit rate α

based on a current error rate ε

of the first agent and a desired level of accuracy λ

;

(c) if the audit rate is higher than a threshold rate, re-queuing the set of audio recordings for transcription by a second remote agent among a plurality of remote agents; and

(d) receiving a second transcription of the set of audio recordings from the second remote agent;

wherein audit rate α

indicates a frequency with which a transcript should be audited;

said current error rate ε

of the first agent represents a probability that the first transcription is incorrect; and

said desired level of accuracy λ

reflects a probability that a transcription is correct.

10. An apparatus for facilitating transcription of an audio recording, with auditing, the apparatus comprising:

an IVR (Interactive Voice Response) unit configured to record verbal input during a telephone call from a caller;

a database configured to store a set of audio recordings of the verbal input;

communication connections to a plurality of agents for transcribing audio recordings;

a web server configured to serve the set of audio recordings to one or more of the agents; and

audit logic configured to determine whether to audit a transcription of the set of audio recordings performed by one agent by calculating an audit rate α

based on a current error rate ε

of the one agent and a desired level of accuracy λ and

, if the audit rate is higher than a threshold rate, having another agent transcribe the set of audio recordings;

wherein audit rate α

indicates a frequency with which a transcript should be audited;

said current error rate ε

of the first agent represents a probability that the first transcription is incorrect; and

said desired level of accuracy λ

reflects a probability that a transcription is correct.

11. The apparatus of claim 10, wherein said audit logic is configured to cause the set of audio recordings to be transcribed by different agents up to a threshold number of times.

12. The apparatus of claim 10, wherein said audit rate α

is calculated as α

≧

(λ

+ε

−

1)/(ε

−

ε

ⁿ);

wherein n>

1 and represents a maximum number of transcriptions of the set of audio recordings to perform.

13. The apparatus of claim 10, wherein said audit logic determines whether to have the set of audio recordings transcribed by another agent by flipping a weighted coin configured based on said audit rate α

.

14. The apparatus of claim 10, further comprising:

an internal server configured to store the set of audio recordings in the database for the IVR unit.

15. The apparatus of claim 10, wherein said web server is further configured to serve the agents a form designed to receive the agents'"'"' transcriptions.