2 Speaker identification
2.4 Classification/decision taking algorithm

After the feature extraction and possible normalization/ compensation phase (will be presented in the next section) a classification must be used to decide which user (features or models) are the closes to the unknown one. Still it is possible to reject any if the recorded match/ confidence is too low. There are several successful classification techniques that differ in their complexity, way how they work and what they assume upon the processed data. These methods are classified into several major categories each with its pros and cons as follows:

It was discovered that it is beneficial to use general models of universal speakers that have been created by training samples from many speakers.