Back to results
US 11,380,304 B1Granted
Generation of alternate representions of utterances
AssigneeAmazon Technologies, Inc.
InventorsPragaash Ponnusamy, Alireza Roshan Ghias, Chenlei Guo
Life of this patent
Estimating the term…
Abstract
A system is provided for handling errors during automatic speech recognition by processing a potentially defective utterance to determine an alternative, potentially successful utterance. The system processes an ASR hypothesis, using a probabilistic graph, to determine a likelihood that it will result in an error. Using the probabilistic graph, the system determines an alternate utterance.
Claims20
A computer-implemented method comprising:
- receiving, from a user device, first audio data representing a first utterance;
- performing automatic speech recognition (ASR) processing on the first audio data to generate first ASR output data representing the first utterance;
- determining first probability data indicating a first likelihood that a first system response to the first utterance results in an undesired response;
- determining, using a Markov Chain probabilistic graph, second ASR output data corresponding to a second utterance based on second probability data indicating a second likelihood that a second system response to the second utterance results in a desired response, the second utterance corresponding to a rephrasing of the first utterance;
- performing NLU processing on the second ASR output data, instead of the first ASR output data to determine a first intent representing the second utterance;
- and
- generating first output data based at least in part on the first intent.
4 more elements
A computer-implemented method comprising:
- receiving a first representation of a first natural language user input;
- determining first probability data indicating a first likelihood that a first system response to the first natural language user input results in an undesired response;
- determining, using a Markov Chain probabilistic graph, a second representation of a second natural language user input based on second probability data indicating a second likelihood that a second system response to the second natural language user input results in a desired response, the second natural language user input corresponding to a rephrasing of the first natural language user input;
- processing the second representation to determine a first intent representing the second natural language user input;
- and
- generating first output data based at least in part on the first intent.
3 more elements
A system, comprising:
- at least one processor;
- and
- at least one memory comprising instructions that, when executed by the at least one processor, cause the system to:
- receive a first representation of a first natural language user input;
- determine first probability data indicating a first likelihood that a first system response to the first natural language user input results in an undesired response;
- determine, using a Markov Chain probabilistic graph, a second representation of a second natural language user input based on second probability data indicating a second likelihood that a second system response to the second natural language user input results in a desired response, the second natural language user input corresponding to a rephrasing of the first natural language user input;
- process the second representation to determine a first intent representing the second natural language user input;
- and
- generate first output data based at least in part on the first intent.
6 more elements
Dependencies are inferred from the claim text.