Two Statistical Feature Evaluation Techniques Applied to Speaker Identification

The problem of identifying people based solely upon samples of their speech is viewed as a statistical pattern classification problem, emphasizing the portion of the process in which an informative, concise set of features is extracted from the speech signal. This work takes into account both the limited amount of data available in a real application and the statistical dependence among the various proposed features. In addition, the results of feature evaluation should apply to speakers not involved in the evaluation set.