Effect of VoIP on prosodic features for speaker verification

摘要

With the increased use of voice over IP (VoIP) telephony, remote speaker verification systems has gained importance. In this work, focus is on prosody based speaker verification systems where the test speech is sent over the IP networks and the claim is verified on a remote server. When the test data is sent over IP network, the performance of speaker verification system gets drastically reduced due to the compression effect introduced by the codecs, packet loss and noise. In this paper, a study of the effect of VoIP on prosodic features is done. Among the nine prosodic features namely - four weights of Legendre polynomial, jitter, delta energy, shimmer, phrase duration and voiced duration, those features which vary the most in the case of VoIP is found. It is done by capturing the variations of different features over the syllables. Also normal speech and VoIP is applied to the prosody based speaker verification system and the performance is evaluated.