Communication issue detection using evaluation of multiple machine learning models
Techniques are provided for evaluating multiple machine learning models to identify issues with a communication. One method comprises applying an audio signal associated with a communication to at least two of: (i) a trigger word analysis module that evaluates contextual information to determine if a trigger word is detected in the audio signal; (ii) an audio activity pattern analysis module that determines if a silence pattern anomaly is detected; and (iii) a communication application analysis module that evaluates features provided by a communication application relative to applicable thresholds; and combining results of the at least two of the trigger word analysis module, the audio activity pattern analysis module and the communication application analysis module to identify a communication issue. The combining may evaluate an accuracy of the trigger word analysis module, the audio activity pattern analysis module and/or the communication application analysis module to combine the results.
1. A method, comprising:
applying, by a communication issue detector, a representation of an audio signal associated with a communication to:
(i) a trigger word analysis module that determines if one or more trigger words are detected in the audio signal from the audio signal using a trained trigger word detection model that is trained using (a) a set of trigger words indicative of a technical device issue with one or more devices associated with the communication and (b) contextual information comprising one or more of a nearby additional trigger word feature, a time-in-meeting of trigger word feature and a before/after silence pattern feature, relative to a user-provided feedback score; and
(ii) an audio activity pattern analysis module that determines if a silence pattern anomaly is detected in the audio signal using a trained audio activity model, wherein the audio activity pattern analysis module evaluates a length of an audio activity portion of the audio signal relative to a length of a silence portion of the audio signal to identify the silence pattern anomaly, wherein the silence pattern anomaly is indicative of a technical device issue with one or more devices associated with the communication;
combining results of the trigger word analysis module and the audio activity pattern analysis module to identify a communication issue for the communication; and
implementing one or more remedial actions responsive to the identification of the communication issue;
wherein the method is performed by at least one processing device comprising a processor coupled to a memory.
2. The method of claim 1 , wherein the representation of the audio signal comprises one or more spectrograms associated with the communication.
3. The method of claim 1 , wherein the applying the audio signal to the trigger word analysis module further comprises evaluating a relevance score generated by the trained trigger word detection model.
4. The method of claim 1 , wherein the trained trigger word detection model is further trained using a plurality of additional words and a plurality of background samples.
5. The method of claim 1 , wherein the combining employs an ensemble model that combines the results to identify the communication issue for the communication.
6. The method of claim 1 , wherein the combining evaluates an accuracy of the trigger word analysis module and the audio activity pattern analysis module to combine the results.
7. The method of claim 1 , further comprising applying the representation of the audio signal associated with a communication to a communication application analysis module that evaluates one or more features provided by a communication application relative to one or more thresholds, wherein the communication application is provided by a different provider than a provider of the communication issue detector.
8. The method of claim 7 , wherein the features provided by the communication application comprise one or more of an audio device not found feature, a number of screen share sessions feature, a number of connection attempts feature and a poor connection events feature.
9. An apparatus comprising:
at least one processing device comprising a processor coupled to a memory;
the at least one processing device being configured to implement the following steps:
applying, by a communication issue detector, a representation of an audio signal associated with a communication to:
(i) a trigger word analysis module that determines if one or more trigger words are detected in the audio signal from the audio signal using a trained trigger word detection model that is trained using (a) a set of trigger words indicative of a technical device issue with one or more devices associated with the communication and (b) contextual information comprising one or more of a nearby additional trigger word feature, a time-in-meeting of trigger word feature and a before/after silence pattern feature, relative to a user-provided feedback score; and
(ii) an audio activity pattern analysis module that determines if a silence pattern anomaly is detected in the audio signal using a trained audio activity model, wherein the audio activity pattern analysis module evaluates a length of an audio activity portion of the audio signal relative to a length of a silence portion of the audio signal to identify the silence pattern anomaly, wherein the silence pattern anomaly is indicative of a technical device issue with one or more devices associated with the communication;
combining results of the trigger word analysis module and the audio activity pattern analysis module to identify a communication issue for the communication; and
implementing one or more remedial actions responsive to the identification of the communication issue.
10. The apparatus of claim 9 , wherein the applying the audio signal to the trigger word analysis module further comprises evaluating a relevance score generated by the trained trigger word detection model.
11. The apparatus of claim 9 , wherein the trained trigger word detection model is trained using a set of trigger words, a plurality of additional words and a plurality of background samples.
12. The apparatus of claim 9 , wherein the combining employs an ensemble model that combines the results to identify the communication issue for the communication.
13. The apparatus of claim 9 , wherein the combining evaluates an accuracy of the trigger word analysis module and the audio activity pattern analysis module to combine the results.
14. The apparatus of claim 9 , wherein the representation of the audio signal comprises one or more spectrograms associated with the communication.
15. The apparatus of claim 9 , further comprising applying the representation of the audio signal associated with a communication to a communication application analysis module that evaluates one or more features provided by a communication application relative to one or more thresholds, wherein the communication application is provided by a different provider than a provider of the communication issue detector.
16. A non-transitory processor-readable storage medium having stored therein program code of one or more software programs, wherein the program code when executed by at least one processing device causes the at least one processing device to perform the following steps:
applying, by a communication issue detector, a representation of an audio signal associated with a communication to:
(i) a trigger word analysis module that determines if one or more trigger words are detected in the audio signal from the audio signal using a trained trigger word detection model that is trained using (a) a set of trigger words indicative of a technical device issue with one or more devices associated with the communication and (b) contextual information comprising one or more of a nearby additional trigger word feature, a time-in-meeting of trigger word feature and a before/after silence pattern feature, relative to a user-provided feedback score; and
(ii) an audio activity pattern analysis module that determines if a silence pattern anomaly is detected in the audio signal using a trained audio activity model, wherein the audio activity pattern analysis module evaluates a length of an audio activity portion of the audio signal relative to a length of a silence portion of the audio signal to identify the silence pattern anomaly, wherein the silence pattern anomaly is indicative of a technical device issue with one or more devices associated with the communication;
combining results of the trigger word analysis module and the audio activity pattern analysis module to identify a communication issue for the communication; and
implementing one or more remedial actions responsive to the identification of the communication issue.
17. The non-transitory processor-readable storage medium of claim 16 , wherein the applying the audio signal to the trigger word analysis module further comprises evaluating a relevance score generated by the trained trigger word detection model.
18. The non-transitory processor-readable storage medium of claim 16 , wherein the combining employs an ensemble model that combines the results to identify the communication issue for the communication.
19. The non-transitory processor-readable storage medium of claim 16 , wherein the combining evaluates an accuracy of the trigger word analysis module and the audio activity pattern analysis module to combine the results.
20. The non-transitory processor-readable storage medium of claim 16 , further comprising applying the representation of the audio signal associated with a communication to a communication application analysis module that evaluates one or more features provided by a communication application relative to one or more thresholds, wherein the communication application is provided by a different provider than a provider of the communication issue detector.