Automated speech recognition proxy system for natural language understanding
View Patent ↗An interactive response system mixes HSR subsystems with ASR subsystems to facilitate overall capability of user interfaces. The system permits imperfect ASR subsystems to nonetheless relieve burden on HSR subsystems. An ASR proxy is used to implement an IVR system, and the proxy dynamically selects one or more recognizers from a language model and a human agent to recognize user input. Selection of the one or more recognizers is based on factors such as confidence thresholds of the ASRs and availability of human resources for HSRs.
1. A computer-implemented system for processing a user interaction received from a device of the user over a computer network, the interaction including text requiring recognition before being usable for further computer-implemented processing, the system comprising:
an application configured to provide the text;
a recognition decision engine configured to receive the text for recognition, the recognition decision engine dynamically selecting one or more recognizers from:
a language model; and
a human agent using a device located at a location remote from the computer-implemented system; and
a results decision engine configured to provide a recognition result using the selected recognizer.
2. The system of claim 1 , further comprising a system status subsystem operably connected to the recognition decision engine, the recognition decision engine taking as input system load information from the system status subsystem for use in the dynamically selecting.
3. The system of claim 1 , wherein a subset of the one or more recognizers is configured to provide a confidence metric to the recognition decision engine, the recognition decision engine using the confidence metric in the dynamically selecting.
4. The system of claim 3 , wherein the confidence metric includes a threshold, the threshold varying based on resource availability.
5. The system of claim 1 , wherein the recognition decision engine is configured to favor selection of the language model relative to the human agent based on recognition cost factors.
6. The system of claim 1 , wherein the recognition decision engine is configured to favor selection of the language model relative to the human agent based on human resource availability factors.
7. The system of claim 1 , wherein the results decision engine is configured to update confidence thresholds associated with a first one of the recognizers responsive to agreement of results between the first one of the recognizers and a second one of the recognizers.
8. The system of claim 1 , wherein the recognition decision engine is configured to, responsive to initial results provided by the selected recognizer, make a subsequent selection of a second one of the recognizers, the subsequent selection being made before processing of the text is completed by the selected recognizer.
9. A computer-implemented method performed by a computer-implemented system for processing a user interaction from a device of the user over a computer network, the interaction including text requiring recognition before being usable for further computer-implemented processing, the method comprising:
receiving data representing the text from an application;
dynamically selecting one or more recognizers from:
a language model; and
a human agent using a device located at a location remote from the computer-implemented system; and
providing a recognition result responsive to results of processing by the selected recognizer.
10. The computer-implemented method of claim 9 , wherein said dynamically selecting is responsive to a system load metric.
11. The computer-implemented method of claim 9 , wherein said dynamically selecting is responsive to a confidence metric.
12. The computer-implemented method of claim 9 , wherein the confidence metric includes a threshold, the threshold varying based on resource availability.
13. The computer-implemented method of claim 9 , wherein said dynamically selecting favors selection of the language model relative to the human agent based on recognition cost factors.
14. The computer-implemented method of claim 9 , wherein said dynamically selecting favors selection of the language model relative to the human agent based on human resource availability factors.
15. The computer-implemented method of claim 9 , further comprising updating confidence thresholds associated with a first one of the recognizers responsive to agreement of results between the first one of the recognizers and a second one of the recognizers.
16. The computer-implemented method of claim 9 , further comprising initially choosing the language model and, responsive to initial results provided by the selected recognizer, making a subsequent selection of a second one of the recognizers, the subsequent selection being made before processing of the text is completed by the selected recognizer.
17. A non-transitory computer-readable storage medium storing executable computer program code for processing, by a computer-implemented system, a user interaction received from a device of the user over a computer network, the interaction including text requiring recognition before being usable for further computer-implemented processing, the computer program code comprising instructions for:
receiving data representing the text from an application;
dynamically selecting one or more recognizers from:
a language model; and
a human agent using a device located at a location remote from the computer-implemented system; and
providing a recognition result responsive to results of processing by the selected recognizer.
18. The non-transitory computer-readable storage medium of claim 17 , wherein said dynamically selecting is responsive to a system load metric.
19. The non-transitory computer-readable storage medium of claim 17 , wherein said dynamically selecting is responsive to a confidence metric.
20. The non-transitory computer-readable storage medium of claim 17 , wherein the confidence metric includes a threshold, the threshold varying based on resource availability.