Aspects of Stochastic Process Identification Using Wiener Canonical Forms

摘要

The Wiener Canonical Expansion (WCE) procedure of Bayes' decision rule is considered from a pattern recognition standpoint in the form of a phi machine. This approach provides some insight about the roles which the various stages and parameters of the expansion assume. Experiments conducted previously for speaker discrimination using the WCE procedure were reperformed in order to correct for a processing error. The conclusions of these revised experiments were the same as in the previous work; speaker identification was not realized with any degree of certainty by processing raw speaker data as if it were representative of some stationary random process. (Author)