×

Efficient generation of personalized spoken language understanding models

  • US 10,109,273 B1
  • Filed: 08/29/2013
  • Issued: 10/23/2018
  • Est. Priority Date: 08/29/2013
  • Status: Active Grant
First Claim
Patent Images

1. A system comprising:

  • a data harvesting service comprising one or more server computing devices configured to at least;

    store spoken language understanding personalization data associated with a user, wherein;

    the spoken language understanding personalization data comprises a plurality of separate data sets; and

    each of the plurality of separate data sets is accessible independently from each remaining data set of the plurality of data sets;

    receive a new data set of the spoken language understanding personalization data, wherein;

    the new data set is associated with the user; and

    the new data set comprises activity data received from a plurality of services, wherein the activity data is related to actions initiated by the user that cause one or more of the plurality of services to change user-specific data associated with the user; and

    transmit, to a model generation service, a notification that the new data set associated with the user is available; and

    a model generation service comprising one or more server computing devices configured to at least;

    generate a personal language model, associated with the user, based at least in part on at least a portion of the plurality of separate data sets;

    receive the notification that the new data set associated with the user is available;

    determine a predicted action of the user based at least in part on the activity data of the new data set;

    determine that a word associated with the predicted action is not present in a general language model; and

    in response to determining that the word is not present in the general language model, modify the personal language model based at least partly on a likelihood that the user will use the word; and

    a speech recognition service comprising one or more computing devices configured to at least;

    receive audio data representing a user utterance;

    access the personal language model that is modified; and

    generate speech recognition results using the audio data and the personal language model that is modified.

View all claims
  • 1 Assignment
Timeline View
Assignment View
    ×
    ×