Biometric Security Systems: Speaker Recognition Technology

speaker recognition, speaker verification, speaker identification, biometric security system, hidden Markov model

It has been quite a long time since the first biometric systems were introduced and, until present, they have not become widely used. In spite of this, the importance of the biometric security systems is growing now. Many research groups are working on the development of the individual biometric technologies such as the fingerprint recognition, iris or retina recognition or the speaker recognition. However, not many of them combine these technologies together. Future of the biometric security systems is in the combination of more technologies. This work aims at the technology of the speaker recognition and proposes a solution of its integration into a more complex biometric security system.

There are many speaker recognition systems but none of them is reliable and stable enough to be applied as a standalone security system. In the first part of the dissertation there are described the goals of this works and the state of art of the biometry. Then, description of some methods of the common signal processing and techniques of the pre-processing phase together with feature extraction follows. The feature extraction is a crucial phase of the recognition process. In the next part, some new features focused on the speaker recognition are introduced. The following part aims at the speaker recognition itself. The speaker may be recognised in many ways, but the most suitable one for this task seems to be a method based on the Hidden Markov Models. The next part introduces a design of a complex biometric security system based on the speaker recognition and the fingerprint authentication. A method of acquisition of a unique vector from speaker specific features is outlined as well. If the vector proved to be unique, it would be possible to use it with success in cryptography. In the last part, there are summarised experimental results. Some experiments were performed, their results were summarised and interpreted.