Cepstral analysis in speaker recognition systems

This paper addresses the issues of Automatic Speaker Recognition (ASR) systems. The speech signal in its raw form is characterized by high redundancy, which is why it is necessary to extract specific signal features that allow for an effective description of signal properties important from the point of view of speaker recognition. For this reason, signal parameterization in the recognition process is extremely important. The authors attempted to select an optimal (most discriminating) set of parameters describing the signal based on homomorphic processing methods. The research focused primarily on evaluating the utility of cepstral analysis of the speech signal in recognition systems based on voice samples acquired in digital form.Automatic person recognition is used in all systems that provide restricted services or information, especially when a high degree of security for these systems is required. The growing demand for such systems is driven not only by the development of broadly understood biometric techniques but also by telecommunications and the Internet. An undoubted advantage of person recognition systems based on individual voice characteristics is the fact that (…)

Cepstral analysis in speaker recognition systems – PDF file for download