System and method for efficiently transcribing verbal messages to text

• US 8,583,433 B2
• Filed: 08/06/2012
• Issued: 11/12/2013
• Est. Priority Date: 03/28/2002
• Status: Expired due to Fees

• Current Assignee: Avaya LLC (Avaya Incorporated), Avaya Management L.P. (Avaya Incorporated)
• Original Assignee: Intellisist, Inc. dba Spoken Communications (Avaya Incorporated)
• Inventors: Webb, Mike O., Peterson, Bruce J., Kaseda, Janet S.
• Primary Examiner(s): Shah, Paras D
• Application Number:
  US13/568,065
• Publication Number:
  US 20130035937A1
• Time in Patent Office: 463 Days
• Field of Search: 704/231, 704/235, 704/239, 704/251, 704/254, 704/270, 704/275, 379 8801- 8828, 700 99-111
• US Class Current: 704/235
• CPC Class Codes: G10L 15/04 Segmentation; Word boundary detection
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• 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 2203/2016 Call initiation by network rather than by subscriber
• CPC Class Codes: H04M 2203/4536 Voicemail combined with text-based messaging
• CPC Class Codes: H04M 3/53333 Message receiving aspects

Abstract

A system and method for efficiently transcribing verbal messages to text is provided. Verbal messages are received and at least one of the verbal messages is divided into segments. Automatically recognized text is determined for each of the segments by performing speech recognition and a confidence rating is assigned to the automatically recognized text for each segment. A threshold is applied to the confidence ratings and those segments with confidence ratings that fall below the threshold are identified. The segments that fall below the threshold are assigned to one or more human agents starting with those segments that have the lowest confidence ratings. Transcription from the human agents is received for the segments assigned to that agent. The transcription is assembled with the automatically recognized text of the segments not assigned to the human agents as a text message for the at least one verbal message.

Claims

1. A system for transcribing verbal messages into text, comprising:

verbal messages; and

a processor to execute the following modules, comprising;

a segment module to divide one such verbal message into segments;

a speech recognizer module to determine automatically recognized text for each of the segments and a confidence rating assigned to the automatically recognized text for that segment, wherein the confidence rating comprises a probability that the automatically recognized text is accurate;

a threshold module to apply a threshold to the confidence ratings and identifying those segments with confidence ratings that fall below the threshold;

an assignment module to provide the segments that fall below the threshold to a workbench partial message queue for assigning to one or more human agents starting with those segments that have the lowest confidence ratings, to withhold the segments that are above the threshold from the workbench partial message queue and to automatically output the withheld segments for assembly into a text message for the verbal message;

a receipt module to receive transcription from the human agents for the segments assigned to that human agent; and

an assembly module to assemble the received transcription with the automatically recognized text of the withheld segments in the text message.

2. A system according to claim 1, further comprising:

an identification module to identify a rise in speech recognition performance; and

the assignment module to assign to the human agents only those segments that have the lowest confidence ratings and retaining the automatically recognized text of the remaining segments.

3. A system according to claim 1, further comprising:

a factor assignment module to assign factors to each human agent comprising at least one of quality, fatigue, and performance factors; and

the assignment module to assign the segments that fall below the threshold to the human agents based on the factors.

4. A system according to claim 1, further comprising:

a further assignment module to assign the verbal message to one or more processors that perform the speech recognition based on assignment rules comprising one or more of message content, message type, and priority level of the message.

5. A system according to claim 1, further comprising:

a format identification module to identify common formats within the verbal message;

the speech recognizer to determine automatically recognized text for the common formats; and

a text assembly module to assemble the automatically recognized text of the common formats with the transcription and automatically recognized text of the withheld segments.

6. A system according to claim 5, wherein the common formats correspond to frequently used phrases and sentences.

7. A system according to claim 1, further comprising:

a segment determination module to determine the segments of the verbal message based on one or more of a point in the message where silence is present and after a specified duration of time.

8. A system according to claim 1, further comprising:

a highlight module to highlight the segments of the verbal message to be transcribed by the human agents.

9. A system according to claim 1, wherein the segments that fall below the threshold are assigned to the human agents based on at least one of message rank, agent availability, and message content.

10. A system according to claim 1, wherein the transcription by the human agents comprises at least one of editing the automatically recognized text and manual transcription of the segment.

11. A method for transcribing verbal messages into text, comprising the steps of:

receiving verbal messages;

dividing one such verbal message into segments;

determining, via a processor, automatically recognized text for each of the segments by performing speech recognition and a confidence rating assigned to the automatically recognized text for that segment, wherein the confidence rating comprises a probability that the automatically recognized text is accurate;

applying a threshold to the confidence ratings and identifying those segments with confidence ratings that fall below the threshold;

providing the segments that fall below the threshold to a workbench partial message queue for assigning to one or more human agents starting with those segments that have the lowest confidence ratings;

withholding the segments that are above the threshold from the workbench partial message queue and automatically outputting the withheld segments for assembly into a text message for the verbal message;

receiving transcription from the human agents for the segments assigned to that human agent; and

assembling the received transcription with the automatically recognized text of the withheld segments in the text message.

12. A method according to claim 11, further comprising:

identifying a rise in speech recognition performance; and

assigning to the human agents only those segments that have the lowest confidence ratings and retaining the automatically recognized text of the remaining segments.

13. A method according to claim 11, further comprising:

assigning factors to each human agent comprising at least one of quality, fatigue, and performance factors; and

assigning the segments that fall below the threshold to the human agents based on the factors.

14. A method according to claim 11, further comprising:

assigning the verbal message to one or more processors that perform the speech recognition based on assignment rules comprising one or more of message content, message type, and priority level of the message.

15. A method according to claim 11, further comprising:

identifying common formats within the verbal message;

determining automatically recognized text for the common formats; and

assembling the automatically recognized text of the common formats with the transcription and automatically recognized text of the withheld segments.

16. A method according to claim 15, wherein the common formats correspond to frequently used phrases and sentences.

17. A method according to claim 11, further comprising:

determining the segments of the at least one verbal message based on one or more of a point in the message where silence is present and after a specified duration of time.

18. A method according to claim 11, further comprising:

highlighting the segments of the verbal message to be transcribed by the human agents.

19. A s method according to claim 11, wherein the segments that fall below the threshold are assigned to the human agents based on at least one of message rank, agent availability, and message content.

20. A method according to claim 11, wherein the transcription by the human agents comprises at least one of editing the automatically recognized text and manual transcription of the segment.