The Impact of Telephone Channels on the Accuracy of Automatic Speaker Recognition

Abstract

This paper presents an experimental study on the impact of telephone channels on the accuracy of automatic speaker recognition. Speaker models and the design of the recognizer used in this study are based on Hidden Markov models. In order to simulate telephonequality speech signals, several experimental conditions were introduced taking two control factors into consideration: the type of the applied codec and the probability of transmission errors. In addition, the impact of echo signals – that are often present in Internet telephony – on the accuracy of automatic speaker recognition systems is considered. Finally, the paper provides a brief overview of several methodologies for the adaptation of the recognizer to the expected environmental conditions that may enhance the robustness of the speaker recognizer.

Keywords

• Automatic speaker recognition
• experimental study
• telephone channel
• echo in VoIP
• adaptation