Speaker identification using cepstral analysis

Human voice is characteristic for an individual. The ability to recognize a speaker by his/her voice can be a valuable biometric tool with enormous commercial as well as academic potential. Commercially, it can be utilized for ensuring secure access to any system. Academically, it can shed light on the speech processing abilities of the brain as well as the speech mechanism. In fact, this feature is being used preliminarily, along with other biometrics, including face, iris and fingerprint recognition, for commercial security products. How humans recognize someone from his/her voice has been an elusive ability to define and reproduce artificially. Research is still ongoing in this area. The paper views the speaker identification problem from the perspective of digital signal processing. A solution based on cepstral analysis is investigated and results presented. The fair success rate of the solution argues that the cepstral analysis approach is promising.

1. Introduction

Speaker Recognition or Speaker Verification/Authentication is the term used to imply the automatic process of identifying a given speaker from amongst a number of already registered persons, through the “voiceprint” of the speaker. The voiceprint consists of a set of features which are supposedly unique for the given speaker.