×

Accuracy model for recognition signal processing engines

  • US 7,580,570 B2
  • Filed: 12/09/2003
  • Issued: 08/25/2009
  • Est. Priority Date: 12/09/2003
  • Status: Expired due to Fees
First Claim
Patent Images

1. A method for evaluating a computer learning signal processing engine, comprising:

  • employing at least one processor to execute computer-executable instructions stored on at least one computer-readable medium to perform the following acts;

    identifying a first group of signal sets representing inputs by humans, each signal set of the first group having an associated range of values for a variable corresponding to the first group, the variable being one of a plurality of variables having values characterizing multiple signals to be processed, the plurality of variables including at least a variable characterizing a user scenario in which a signal was generated, the user scenario including at least one of a software application or an operation performed within a software application;

    calculating an accuracy score for each signal set of the first group using the signal processing engine to be evaluated;

    applying weight factors to the accuracy scores for the first group signal sets, each weight factor representing a relative importance of one of the associated ranges of values for the first variable;

    summing weighted accuracy scores for the first group of signal sets to yield a first summed accuracy score;

    identifying additional groups of signal sets, each group having a corresponding variable of the plurality of variables, each signal set of a group having an associated range of values for the corresponding variable;

    calculating accuracy scores for each signal set of each additional group using the signal processing engine to be evaluated;

    applying weight factors to the accuracy scores for the signal sets of the additional groups;

    summing the weighted accuracy scores within each of the additional groups to yield additional summed accuracy scores;

    using summed accuracy scores from at least two separate training sets comprising samples, wherein each separate training set is distinguished by a feature characteristic identified based upon a demographic characteristic associated with a source of the samples, to create one or more signal processing engines to handle multiple applications to one or more new groups of signal sets for which a frequency of the feature characteristic of the separate training sets are known or assumed, by weighting the summed accuracy score associated with each training set according to the frequency and then combining the weighted summed accuracy scores to yield scone-specific accuracy scores;

    applying weights to each of the scope-specific accuracy scores;

    summing the weighted scope-specific accuracy scores to yield a combined accuracy score;

    summing the combined accuracy score with at least one other summed accuracy score relating to a physical attribute of one or more input signals to yield an input accuracy score; and

    summing the input accuracy score with a summed accuracy score for a group of signal sets corresponding to the user scenario variable to yield an overall accuracy score.

View all claims
  • 2 Assignments
Timeline View
Assignment View
    ×
    ×