Back to results
US 8,571,869 B2Granted
Natural language system and method based on unisolated performance metric
AssigneeNuance Communications, Inc.
InventorsSabine Deligne, Yuqing Gao, Vaibhava Goel, Hong-Kwang Kuo, Cheng Wu
Life of this patent
Estimating the term…
Abstract
A natural language business system and method is developed to understand the underlying meaning of a person's speech, such as during a transaction with the business system. The system includes a speech recognition engine, and action classification engine, and a control module. The control module causes the system to execute an inventive method wherein the speech recognition and action classification models may be recursively optimized on an unisolated performance metric that is pertinent to the overall performance of the natural language business system, as opposed to the isolated model-specific criteria previously employed.
Claims30
A method comprising:
- obtaining first data from a plurality of calls routed by a natural language call routing system that uses a first speech recognition model to recognize speech of the plurality of calls and a first action classification model to route calls based on the recognized speech, the first data comprising audio data from the plurality of calls, a first N-best list of word sequences generated by using the first speech recognition model to recognize at least a portion of the audio data and associated call classification data indicating how each of the plurality of calls were routed by the natural language call routing system;
- and
- modifying the first speech recognition model and the first action classification model based at least in part on the first data, including using the first N-best list of word sequences, to obtain a second speech recognition model and a second action classification model;
- and
- modifying the second speech recognition model and the second action classification model based at least in part on the first data to obtain a third speech recognition model and a third action classification model, wherein the modifying comprises modifying the second speech recognition model and the second action classification model by using the first N-best list or a second N-best list of word sequences generated by using the second speech recognition model to recognize at least the portion of the audio data.
2 more elements
An article of manufacture comprising at least one non-transitory computer readable medium containing one or more programs which, when executed by at least one processor, cause the at least one processor to perform a method comprising:
- obtaining first data from a plurality of calls routed by a natural language call routing system that uses a first speech recognition model to recognize speech of the plurality of calls and a first action classification model to route calls based on the recognized speech, first data comprising audio data from the plurality of calls, a first N-best list of word sequences generated by using the first speech recognition model to recognize at least a portion of the audio data and associated call classification data indicating how each of the plurality of calls were routed by the natural language call routing system;
- and
- modifying the first speech recognition model and the first action classification model based at least in part on the first data, including using the first N-best list of word sequences, to obtain a second speech recognition model and a second action classification model;
- and
- modifying the second speech recognition model and the second action classification model based at least in part on the first data to obtain a third speech recognition model and a third action classification model, wherein the modifying comprises modifying the second speech recognition model and the second action classification model by using the first N-best list or a second N-best list of word sequences generated by using the second speech recognition model to recognize at least the portion of the audio data.
2 more elements
A natural language call routing system comprising:
- at least one processor configured to:
- obtain first data from a plurality of calls routed by a natural language call routing system that uses a first speech recognition model to recognize speech of the plurality of calls and a first action classification model to route calls based on the recognized speech, the first data comprising audio data from the plurality of calls, a first N-best list of word sequences generated by using the first speech recognition model to recognize at least a portion of the audio data and associated call classification data indicating how each of the plurality of calls were routed by the natural language call routing system;
- and
- modify the first speech recognition model and the first action classification model based at least in part on the first data, including using the first N-best list of word sequences, to obtain a second speech recognition model and a second action classification model;
- and
- modify the second speech recognition model and the second action classification model based at least in part on the first data to obtain a third speech recognition model and a third action classification model, wherein the modifying comprises modifying the second speech recognition model and the second action classification model by using the first N-best list or a second N-best list of word sequences generated by using the second speech recognition model to recognize at least the portion of the audio data.
3 more elements
A method comprising:
- obtaining, using at least one processor, audio data for a plurality of calls handled by a natural language call routing system,
- automatically obtaining, using the at least one processor, call classification data from the audio data by using a first speech recognition model and a first action classification model;
- and
- modifying, by using the at least one processor, the first speech recognition model and the first action classification model based at least in part on the audio data and the call classification data to obtain a second speech recognition model and a second action classification model;
- and
- modifying the second speech recognition model and the second action classification model based at least in part on the first data to obtain a third speech recognition model and a third action classification model, wherein the modifying comprises modifying the second speech recognition model and the second action classification model by using the first N-best list or a second N-best list of word sequences corresponding to at least the portion of the audio data obtained by using the second speech recognition model.
3 more elements
An article of manufacture comprising at least one non-transitory computer readable medium containing one or more programs which, when executed by at least one processor, cause the at least one processor to perform a method comprising:
- obtaining, using the at least one processor, audio data for a plurality of calls handled by a natural language call routing system,
- automatically obtaining, using the at least one processor, call classification data from the audio data by using a first speech recognition model and a first action classification model;
- and
- modifying, using the at least one processor, the first speech recognition model and the first action classification model based at least in part on the audio data and the call classification data to obtain a second speech recognition model and a second action classification model;
- and
- modifying the second speech recognition model and the second action classification model based at least in part on the first data to obtain a third speech recognition model and a third action classification model, wherein the modifying comprises modifying the second speech recognition model and the second action classification model by using the first N-best list or a second N-best list of word sequences corresponding to at least the portion of the audio data obtained by using the second speech recognition model.
3 more elements
A natural language routing system comprising:
- at least one processor configured to:
- obtain audio data for a plurality of calls handled by a natural language call routing system,
- automatically obtain call classification data from the audio data by using a first speech recognition model and a first action classification model;
- and
- modify the first speech recognition model and the first action classification model based at least in part on the audio data and the call classification data to obtain a second speech recognition model and a second action classification model;
- and
- modify the second speech recognition model and the second action classification model based at least in part on the first data to obtain a third speech recognition model and a third action classification model, wherein the modifying comprises modifying the second speech recognition model and the second action classification model by using the first N-best list or a second N-best list of word sequences corresponding to at least the portion of the audio data obtained by using the second speech recognition model.
4 more elements
Dependencies are inferred from the claim text.