Computer-implemented system and method for transcription error reduction during a live call
A computer-implemented system and method for transcription error reduction during a live call is provided. Speech utterances are transcribed during a live call. A confidence score is assigned to each transcribed speech utterance. One of the transcribed speech utterances with a low confidence score is identified. An attempt is made to identify a pool of related transcribed speech utterances by monitoring other live calls for a predetermined amount of time. Similar transcribed speech utterances with low confidence scores are identified and a group of the similar transcribed speech utterances and the identified transcribed speech utterance is formed. A determination is made as to whether the group includes a predetermined number of transcribed speech utterances upon termination of the predetermined time. If so, a sample of the transcribed speech utterances in the pool is processed to verify a validity of the transcribed speech utterances.
1. A computer-implemented system for transcription error reduction during a live call, comprising:
a processor to execute modules, comprising:
a transcription module configured to transcribe speech utterances from a speaker during a live call;
a scoring module configured to assign a confidence score to each transcribed speech utterance from the live call;
an identification module configured to identify one of the transcribed speech utterances with a low confidence score;
a grouping module configured to attempt to generate a pool of related transcribed speech utterances for the identified transcribed speech utterance, comprising:
a monitor submodule configured to monitor other speech utterances from other live calls for a predetermined amount of time;
a similarity submodule configured to identify similar transcribed speech utterances with low confidence scores from one or more of the other live calls and to form a group comprising the identified similar transcribed speech utterances and the identified transcribed speech utterance;
a group determination submodule configured to determine whether the group includes a predetermined number of transcribed speech utterances upon termination of the predetermined time; and
a group designation submodule configured to designate the group as the pool when the group has at least the predetermined number of transcribed speech utterances within the predetermined time; and
a sample processing module configured to process a sample of the transcribed speech utterances in the pool to verify a validity of the transcribed speech utterances in the sample and pool.
2. A system according to claim 1 , wherein the speech utterances from the live call are transcribed by accessing one or more grammars, selecting a transcribed value for each speech utterance from one of the grammars, and assigning the transcribed speech values to the speech utterances.
3. A system according to claim 2 , further comprising:
a grammar selection module configured to select at least one of the grammars based on at least one characteristic of the speaker.
4. A system according to claim 2 , further comprising:
a message generator configured to generate a message for the live call based on the transcribed speech values of the speech utterances.
5. A system according to claim 4 , further comprising:
a database configured to store the message with a recording of the live call.
6. A system according to claim 1 , further comprising:
a confidence determination module configured to determine the identified transcribed speech utterance as having a low confidence score by applying a threshold to the assigned confidence score and designating the identified transcribed speech utterance as having the low confidence score when the confidence score falls below the threshold.
7. A system according to claim 6 , wherein the low confidence score represents a high likelihood that the transcribed speech utterance incorrectly represents the corresponding speech utterance.
8. A system according to claim 1 , further comprising:
a similarity determination module configured to determine a similarity of the similar transcribed speech utterances and the identified transcribed speech utterance based on one or more similarity metrics.
9. A system according to claim 1 , further comprising:
a threshold determination module configured to set one or more of the predetermined number of transcribed speech utterances and the predetermined time based on at least one of call type, accuracy of transcription required, and type of service agreement signed by the speaker.
10. A system according to claim 9 , wherein the predetermined number of transcribed speech utterances is lower for the live call than for speech utterances from a call recording.
11. A computer-implemented method for transcription error reduction during a live call, comprising:
transcribing speech utterances from a speaker during a live call;
assigning a confidence score to each transcribed speech utterance from the live call;
identifying one of the transcribed speech utterances with a low confidence score;
attempting to generate a pool of related transcribed speech utterances for the identified transcribed speech utterance, comprising:
monitoring other speech utterances from other live calls for a predetermined amount of time;
identifying similar transcribed speech utterances with low confidence scores from one or more of the other live calls and forming a group comprising the similar transcribed speech utterances and the identified transcribed speech utterance;
determining whether the group includes a predetermined number of transcribed speech utterances upon termination of the predetermined time; and
designating the group as the pool when the group has at least the predetermined number of transcribed speech utterances within the predetermined time; and
processing a sample of the transcribed speech utterances in the pool to verify a validity of the transcribed speech utterances in the sample and pool.
12. A method according to claim 11 , wherein the speech utterances from the live call are transcribed, comprising:
accessing one or more grammars;
selecting a transcribed value for each speech utterance from one of the grammars; and
assigning the transcribed speech values to the speech utterances.
13. A method according to claim 12 , further comprising:
selecting at least one of the grammars based on at least one characteristic of the speaker.
14. A method according to claim 12 , further comprising:
generating a message for the live call based on the transcribed speech values of the speech utterances.
15. A method according to claim 14 , further comprising:
storing the message with a recording of the live call.
16. A method according to claim 11 , further comprising:
determining the identified transcribed speech utterance as having a low confidence score, comprising:
applying a threshold to the assigned confidence score; and
designating the identified transcribed speech utterance as having the low confidence score when the confidence score falls below the threshold.
17. A method according to claim 16 , wherein the low confidence score represents a high likelihood that the transcribed speech utterance incorrectly represents the corresponding speech utterance.
18. A method according to claim 11 , further comprising:
determining a similarity of the similar transcribed speech utterances and the identified transcribed speech utterance based on one or more similarity metrics.
19. A method according to claim 11 , further comprising:
setting one or more of the predetermined number of transcribed speech utterances and the predetermined time based on at least one of call type, accuracy of transcription required, and type of service agreement signed by the speaker.
20. A method according to claim 19 , wherein the predetermined number of transcribed speech utterances is lower for the live call than for speech utterances from a call recording.