US 10,650,811 B2Granted

Correction of speech recognition on repetitive queries

AssigneeMicrosoft Technology Licensing, LLC

InventorsMeryem Pinar Donmez Ediz, Ranjitha Gurunath Kulkarni, Shuangyu Chang, Nitin Kamra

Life of this patent

Estimating the term…

Loading patent…

Abstract

Disclosed in various examples are methods, systems, and machine-readable mediums for providing improved computer implemented speech recognition by detecting and correcting speech recognition errors during a speech session. The system recognizes repeated speech commands from a user in a speech session that are similar or identical to each other. To correct these repeated errors, the system creates a customized language model that is then utilized by the language modeler to produce a refined prediction of the meaning of the repeated speech commands. The custom language model may comprise clusters of similar past predictions of speech commands from the speech session of the user.

Claims20

Dependent claims

A computing device for speech recognition, the computing device comprising:

a computer processor;
a memory device coupled to the computer processor and comprising instructions, which when performed by the computer processor, causes the computer processor to perform operations comprising:
retrieving a plurality of ranked predictions of a meaning of one or more speech data segments, the one or more speech data segments provided for a particular session of a particular user of a speech recognition application;
clustering the plurality of ranked predictions into one or more clusters based upon a similarity of the ranked predictions;
creating a custom language model for at least one cluster of the one or more clusters based upon the ranked predictions in the at least one cluster;
applying the custom language model in a language modeling process to produce a second plurality of ranked predictions of a meaning of at least one segment of the one or more speech data segments;
and
utilizing the second plurality of ranked predictions to cause a response to the particular user within the particular session.

5 more elements

A computer-implemented method of speech recognition, the method comprising, using one or more computer processors:

retrieving a plurality of ranked predictions of a meaning of one or more speech data segments, the one or more speech data segments provided for a particular session of a particular user of a speech recognition application;
clustering the plurality of ranked predictions into one or more clusters based upon a similarity of the ranked predictions;
creating a custom language model for at least one cluster of the one or more clusters based upon the ranked predictions in the at least one cluster;
applying the custom language model in a language modeling process to produce a second plurality of ranked predictions of a meaning of at least one segment of the one or more speech data segments;
and
utilizing the second plurality of ranked predictions to cause a response to the particular user within the particular session.

3 more elements

A machine-readable medium comprising instructions, which when performed by a machine, causes the machine to perform operations comprising:

retrieving a plurality of ranked predictions of a meaning of one or more speech data segments, the one or more speech data segments provided for a particular session of a particular user of a speech recognition application;
clustering the plurality of ranked predictions into one or more clusters based upon a similarity of the ranked predictions;
creating a custom language model for at least one cluster of the ons:or more clusters based upon the ranked predictions in the at least one cluster;
applying the custom language model in a language modeling process to produce a second plurality of ranked predictions of a meaning of at least one segment of the one or more speech data segments;
and
utilizing the second plurality of ranked predictions to cause a response to the particular user within the particular session.

3 more elements

Dependencies are inferred from the claim text.