Influence of specific voip transmission conditions on speaker recognition problem

Abstract

The paper presents the problem of signal degradation in packet-based voice transmission and its influence on the voice recognition correctness. The Internet is evolving into universal communication network which carries all types of traffic including data, video and voice. Among them the Internet telephony, namely VoIP is going to be an application of a great importance and that is why it is so important to assess how specific conditions and distortions of the Internet transmission (speech coding and most of all packet loss and delay) can influence speaker recognition problem. The Gaussian Mixture Models classification, the feature extraction, the Internet speech transmission standards and the signal degradation methodology applied in the tested system were overviewed. The experiments carried out for two most commonly applied encoders (G.711 and G.723) and three network conditions (poor, average and with no packet loss) revealed a minor significance of the packet loss problem in the tested text-independent system.