USA As Represented by Secretary of the Navy

Tod E Luginbuhl

Roy L Streit

Stephen G Greineder

Michael L Rosseau

A system and method for ranking features by exploiting their relationship the Fisher projection space. The system ranks n features in a feature set using a design set comprising exemplars from each of M possible event classes of an associated feature-based classification system. A training set is created by randomly selecting exemplars from each of the M classes in the design set. A 'smoothed' Fisher projection space for the training set is created by replacing the sample means and the within-class sample covariance matrix normally used in deriving a Fisher projection space with expressions for the mean vectors and covariance matrices derived from event class probability density function estimates. The angle between a given feature and the smoothed Fisher projection space is calculated for each feature in the feature set, and the features are then ordered by increasing numerical size of this angle. The system produces a reduced feature set by eliminating those features which are not important for classification based on the linear ranking of the features.

System and method for feature set reduction

A method for training a speech recognizer in a speech recognition system is described. The method of the present invention comprises the steps of providing a data base containing acoustic speech units, generating a homoscedastic hidden Markov model from the acoustic speech units in the data base, and loading the homoscedastic hidden Markov model into the speech recognizer. The hidden Markov model loaded into the speech recognizer has a single covariance matrix which represents the tied covariance matrix of every Gaussian probability density function PDF for every state of every hidden Markov model structure in the homoscedastic hidden Markov model.

Growing community of inventors

Waterford, CT, United States of America

Tod E Luginbuhl