Generally speaking the task of speaker identification it to automatically decide who the unknown speech sample belongs to. The decision is based on a set of users stored in a database during a training phase. However if the observed decision confidence is too low the system may not recognized anybody.
As there is a fixed set of user who may be recognized at a time, this task is often called as a closed group problem. Speaker identification has been under serious scientific investigation for over 40 years and there is still and even grooving scientific effort going on. With the arrival of new and wide spread technologies it is finding grooving applications in many areas, just to mention few of them:
Speaker identification is quite a tricky problem because of many reasons. Just mention some of them:
However using speech as biometric signals has following advantages:
There are many applications for speaker identification that exhibit different levels of complexity, requirements, confidentiality, response time, etc. Thus we distinguish several major classes of identification systems:
The above mentioned statements and ideas can be further unfolded into many areas and concept spreading over several branches of science. Thus for more thorough introduction to speaker identification problem please refer to e.g. [6].