Computer-implemented system and method for reducing voice transcription error
View Patent ↗A computer-implemented system and method for reducing voice transcription error is provided. Audio data is received during a call and parsed into utterances. A transcribed value and confidence score are assigned to each utterance. Those utterances with low confidence scores are identified as questionable utterances. One of the questionable utterances is selected from the audio data and a pool of questionable utterances is formed. The pool includes the selected questionable utterance and other questionable utterances, from other calls, that are similar to the selected questionable utterance. A size threshold is applied to the pool of questionable utterances. All the questionable utterances in the pool are provided to at least one human transcriber when the pool fails to satisfy the size threshold. Further transcribed values are received from the transcriber for each of the questionable utterances in the pool.
1. A computer-implemented system for reducing voice transcription error, comprising:
a mobile device configured to collect audio during a call, and configured to generate audio data from the collected audio; and
a server comprising a central processing unit, memory, an input port that receives the audio data and an output port, wherein the central processing unit is configured to;
parse the audio data from the mobile device into utterances;
assign a transcribed value and confidence score to each utterance;
identify utterances with low confidence scores as questionable utterances;
select one of the questionable utterances and form a pool of questionable utterances comprising the selected questionable utterance and further comprising other questionable utterances, from other calls, that have transcribed values similar to the transcribed value of the selected questionable utterance;
apply a size threshold to the pool of questionable utterances;
provide each questionable utterance in the pool to a human transcriber when the pool fails to satisfy the size threshold; and
receive from the human transcriber a further transcribed value for each of the questionable utterances in the pool.
2. A system according to claim 1 , further comprising:
a message generator to generate for the call a transcribed message comprising:
the transcribed values for the utterances not identified as questionable; and
one of: the further transcribed value, or the transcribed value assigned to the selected questionable utterance.
3. A system according to claim 2 , wherein the transcribed message is associated with message details comprising at least one of a message identifier, date, time, sender, recipient, identification of related messages, and time required to transcribe the message.
4. A system according to claim 2 , wherein the transcribed message includes the further transcribed value when the further transcribed value is different than the transcribed value assigned to the selected questionable utterance.
5. A system according to claim 1 , wherein the central processing unit is further configured to:
determine a similarity between the selected questionable utterance and the other questionable utterances in the pool, wherein the similarity is based on one or more of the confidence scores, a range of similarity shared between the assigned transcribed values, and characteristics of the call.
6. A system according to claim 1 , wherein the central processing unit is further configured to:
determine that the similar transcribed values of the other questionable utterances and the assigned transcribed value of the selected questionable utterance are similar when the similar transcribed values are the same as the assigned transcribed value of the selected questionable utterance.
7. A system according to claim 1 , wherein the central processing unit is further configured to:
determine that the similar transcribed values of the other questionable utterances and the assigned transcribed value of the selected questionable utterance are similar when the confidence score assigned to the selected questionable utterance satisfies at least one of a confidence score threshold and a predetermined range of confidence, and when a confidence score associated with each of the similar transcribed values satisfies at least one of the confidence score threshold and the predetermined range of confidence.
8. A system according to claim 1 , wherein the central processing unit is further configured to:
determine that the similar transcribed values of the other questionable utterances and the assigned transcribed value of the selected questionable utterance are similar when the similar transcribed values and the assigned transcribed value of the selected questionable utterance are different, and when the similar transcribed values and the assigned transcribed value of the selected questionable utterance both fall within a predetermined range of transcribed values.
9. A system according to claim 1 , wherein the central processing unit is further configured to:
determine based on a desired accuracy of transcription, the size threshold.
10. A computer-implemented method for reducing voice transcription error, comprising:
collecting, via a mobile device, audio during a call, and generating audio data from the collected audio;
gathering via a server the audio data from the mobile device, wherein the server comprises a central processing unit, memory, an input port to receive the audio data, and an output port;
parsing the audio data from the mobile device into utterances;
assigning a transcribed value and confidence score to each utterance;
identifying utterances with low confidence scores as questionable utterances;
selecting one of the questionable utterances from the audio data and forming a pool of questionable utterances comprising the selected questionable utterance and further comprising other questionable utterances, from other calls, that have transcribed values similar to the transcribed value of the selected questionable utterance;
applying a size threshold to the pool of questionable utterances;
providing each questionable utterance in the pool to a human transcriber when the pool fails to satisfy the size threshold; and
receiving from the human transcriber a further transcribed value for each of the questionable utterances in the pool.
11. A method according to claim 10 , further comprising:
generating for the call a transcribed message, the transcribed message comprising the transcribed values for the utterances not identified as questionable and further comprising one of: the further transcribed value, or the transcribed value assigned to the selected questionable utterance.
12. A method according to claim 11 , wherein the transcribed message is associated with message details comprising at least one of a message identifier, date, time, sender, recipient, identification of related messages, and time required to transcribe the message.
13. A method according to claim 11 , wherein the transcribed message includes the further transcribed value when the further transcribed value is different than the transcribed value assigned to the selected questionable utterance.
14. A method according to claim 10 , further comprising:
determining a similarity between the selected questionable utterance and the other questionable utterances in the pool, wherein the similarity is based on one or more of the confidence scores, a range of similarity shared between the assigned transcribed values, and characteristics of the call.
15. A method according to claim 10 , further comprising:
determining that the similar transcribed values of the other questionable utterances and the assigned transcribed value of the selected questionable utterance are similar when the similar transcribed values are the same as the assigned transcribed value of the selected questionable utterance.
16. A method according to claim 15 , further comprising:
determining that the similar transcribed values of the other questionable utterances and the assigned transcribed value of the selected questionable utterance are similar when the confidence score assigned to the selected questionable utterance satisfies at least one of a confidence score threshold and a predetermined range of confidence, and when a confidence score associated with each of the similar transcribed values satisfies at least one of the confidence score threshold and the predetermined range of confidence.
17. A method according to claim 10 , further comprising:
determining that the similar transcribed values of the other questionable utterances and the assigned transcribed value of the selected questionable utterance are similar when the similar transcribed values and the assigned transcribed value of the selected questionable utterance are different, and when the similar transcribed values and the assigned transcribed value of the selected questionable utterance both fall within a predetermined range of transcribed values.
18. A method according to claim 10 , further comprising:
determining based on a desired accuracy of transcription of the data values for the call the size threshold.