Speaker identification is a part of a broader concept known as speaker recognition. It encompasses two important and in a way similar but still different tasks, namely: speaker identification and speaker verification. The first one basically points to an assignment of automatically deciding who the tested voice sample belongs to from a set of users stored in a database during an enrolment (training) phase.
Optionally if the confidence of a final decision is too low nobody is recognized. This task is often referred to as a closed group problem, as there is fixed set of user who may be recognized.
On the other hand verification process evaluates whether the tested individual is the one who he or she claims to be.
As there are many users (possibly 7 billions of living people) it is impossible to have measured characteristic or model for everyone. This task is therefore referred to as open group problem. In this situation model of a general speaker is vital to determine proper acceptance/rejection thresholds.
The speaker recognition is quite a difficult problem because of many reasons mentioned later in the text and has been under serious scientific investigation for over 40 years by many research teams. As there are emerging new and accessible technologies it is finding rapidly grooving application in many areas, just to mention few of them:
Because of the wide spectrum of problems that must be tackled with there are many solutions and common techniques related to speaker identification problem. Those can be classified into 3 main groups:
Finally the task of speaker identification is divided into two major groups i.e. text dependent and text independent problem. In the first group the identification process doesn’t assume any specific text while in the later class the systems require precise text to be uttered. Obviously text dependent system reaches higher accuracies.