Granted Patent B2

US 12,333,473 · App. 18/364,393 · Granted Jun 17, 2025

Inventors: Jithendra Vepa (Bangalore, IN); Jason Turpin (San Francisco, CA); Ayush Kumar (Ranchi, IN); Amrit Dhaliwal (San Francisco, CA); Akshay Kalyani Kore (Bangalore, IN)

Assignee: Observe.AI, Inc.

G06Q10/06398G06F9/451G06F40/284G06F40/295G10L17/22

Abstract

Claims (69)

1. A system, comprising:

a memory; and

a processor coupled to the memory and configured to:

train at least one machine learning model to detect signals, wherein the at least one machine learning model includes a machine learning model that is trained with training data comprising text that is annotated with correct question classifications;

use the at least one machine learning model to detect a plurality of signals associated with an interaction among two or more speaker roles, wherein a signal comprises a text-based signal or an audio-based signal within a respective audio or text stream associated with the two or more speaker roles;

combine two or more signals of the plurality of signals using a prescribed set of operators into a combined signal, wherein the prescribed set of operators are defined by a set of interaction processing configuration information;

determine a recommended event evaluation result based at least in part on whether a criterion associated with an event has been met with respect to the interaction based at least in part on the combined signal;

output, at a user interface:

one or more of audio portions and text portions of the interaction corresponding to the two or more signals that were detected from the interaction and that were combined into the combined signal; and

the recommended event evaluation result corresponding to the event; and

receive a user feedback comprising a correction to the recommended event evaluation result corresponding to the event; and

retrain the at least one machine learning model based at least in part on the correction to improve subsequent signal detection from subsequent interactions.

2. The system of claim 1 , wherein the interaction comprises a first audio stream and a second audio stream, wherein each of the first audio stream and the second audio stream corresponds to respective speech by the two or more speaker roles.

3. The system of claim 2 , wherein the processor is configured to generate a first text stream from the first audio stream and generate a second text stream from the second audio stream.

4. The system of claim 3 , wherein to use the at least one machine learning model to detect the plurality of signals from one or more of the following: the first audio stream, the second audio stream, the first text stream, and the second text stream.

5. The system of claim 3 , wherein to use the at least one machine learning model to detect the plurality of signals from one or more of the following: the first audio stream, the second audio stream, the first text stream, and the second text stream, and wherein to detect the plurality of signals in the interaction includes to determine an entity question and a provided answer signal, including to:

identify the entity question in the first text stream associated with a first speaker role;

determine a first set of temporal information associated with the entity question in the first text stream;

use the first set of temporal information to determine a candidate answer within the second text stream associated with a second speaker role, wherein the candidate answer comprises text that is associated with a second set of temporal information, wherein the second set of temporal information is later than the first set of temporal information;

determine that the candidate answer does correspond to the entity question; and

detect that the entity question and a provided answer are present in the interaction.

6. The system of claim 3 , wherein to use the at least one machine learning model to detect the plurality of signals from one or more of the following: the first audio stream, the second audio stream, the first text stream, and the second text stream, and wherein to detect the plurality of signals in the interaction includes to determine a sympathetic exchange signal, including to:

identify a predetermined keyword in the first text stream associated with a first speaker role;

determine a first set of temporal information associated with the predetermined keyword in the first text stream;

use the first set of temporal information to determine a sympathetic expression from the first text stream;

use the first set of temporal information to determine a response within the second text stream associated with a second speaker role, wherein the response comprises text that is associated with a second set of temporal information, wherein the second set of temporal information is later than the first set of temporal information;

determine that the sympathetic expression and the response form a sympathetic exchange; and

detect that the sympathetic exchange is present in the interaction.

7. The system of claim 1 , wherein the interaction comprises one or more of the following: a recorded phone call, a chat log, a set of emails, and a recorded video.

8. The system of claim 1 , wherein the user interface comprises a first user interface, and wherein the processor is further configured to obtain the set of interaction processing configuration information associated with a specified objective including to:

receive, via a second user interface, the specified objective;

present, via the second user interface, groups of signals associated with the specified objective;

receive, via the second user interface, one or more edits to the groups of signals;

receive, via the second user interface, the prescribed set of operators to use to combine a respective group of signals; and

receive, via the second user interface, event criteria comprising mappings between combined signals and recommended event evaluation results.

9. The system of claim 8 , wherein to obtain the set of interaction processing configuration information associated with the specified objective further includes to receive a set of weights to use to combine the respective group of signals.

10. The system of claim 1 , wherein to detect the plurality of signals in the interaction includes to determine one or more of the following: an articulation signal, a speaking rate signal, an intent signal, a key phrase signal, a dialogue indicating a key phrase plus action signal, a dialogue prompting a negative response signal, and a providing correction information signal.

11. The system of claim 1 , wherein the correction to the recommended event evaluation result comprises a selected evaluation result that is different from the recommended event evaluation result.

12. The system of claim 1 , wherein the prescribed set of operators comprises one or more of the following: AND, OR, NEAR, DIALOG AND, LEFT, and RIGHT.

13. A method, comprising:

training at least one machine learning model to detect signals, wherein the at least one machine learning model includes a machine learning model that is trained with training data comprising text that is annotated with correct question classifications;

using the at least one machine learning model to detect a plurality of signals associated with an interaction among two or more speaker roles, wherein a signal comprises a text-based signal or an audio-based signal within a respective audio or text stream associated with the two or more speaker roles;

combining, using one or more processors, two or more signals of the plurality of signals using a prescribed set of operators into a combined signal, wherein the prescribed set of operators are defined by a set of interaction processing configuration information;

determining a recommended event evaluation result based at least in part on whether a criterion associated with an event has been met with respect to the interaction based at least in part on the combined signal;

outputting, at a user interface:

one or more of audio portions and text portions of the interaction corresponding to the two or more signals that were detected from the interaction and that were combined into the combined signal; and

the recommended event evaluation result corresponding to the event; and

receiving a user feedback comprising a correction to the recommended event evaluation result corresponding to the event; and

retraining the at least one machine learning model based at least in part on the correction to improve subsequent signal detection from subsequent interactions.

14. The method of claim 13 , wherein the interaction comprises a first audio stream and a second audio stream, wherein each of the first audio stream and the second audio stream corresponds to respective speech by the two or more speaker roles.

15. The method of claim 14 , further comprising generating a first text stream from the first audio stream and generating a second text stream from the second audio stream.

16. The method of claim 15 , wherein using the at least one machine learning model to detect the plurality of signals from one or more of the following: the first audio stream, the second audio stream, the first text stream, and the second text stream.

17. The method of claim 16 , wherein to detect the plurality of signals in the interaction includes determining an entity question and a provided answer signal, including:

identifying the entity question in the first text stream associated with a first speaker role;

determining a first set of temporal information associated with the entity question in the first text stream;

using the first set of temporal information to determine a candidate answer within the second text stream associated with a second speaker role, wherein the candidate answer comprises text that is associated with a second set of temporal information, wherein the second set of temporal information is later than the first set of temporal information;

determining that the candidate answer does correspond to the entity question; and

detecting that the entity question and a provided answer are present in the interaction.

18. The method of claim 13 , wherein the correction to the recommended event evaluation result comprises a selected evaluation result that is different from the recommended event evaluation result.

19. A computer program product, the computer program product comprising a non-transitory computer readable storage medium and comprising computer instructions for:

training at least one machine learning model to detect signals, wherein the at least one machine learning model includes a machine learning model that is trained with training data comprising text that is annotated with correct question classifications;

using the at least one machine learning model to detect a plurality of signals associated with an interaction among two or more speaker roles, wherein a signal comprises a text-based signal or an audio-based signal within a respective audio or text stream associated with the two or more speaker roles;

combining two or more signals of the plurality of signals using a prescribed set of operators into a combined signal, wherein the prescribed set of operators are defined by a set of interaction processing configuration information;

determining a recommended event evaluation result based at least in part on whether a criterion associated with an event has been met with respect to the interaction based at least in part on the combined signal;

outputting, at a user interface:

one or more of audio portions and text portions of the interaction corresponding to the two or more signals that were detected from the interaction and that were combined into the combined signal; and

the recommended event evaluation result corresponding to an event; and

receiving a user feedback comprising a correction to the recommended event evaluation result corresponding to the event; and

retraining the at least one machine learning model based at least in part on the correction to improve subsequent signal detection from subsequent interactions.

Continuity (3)

References Cited (21)