Speaker Verification by Human Listeners over Several Speech Transmission Systems

Although a great deal has been learned about how speakers are verified, both by humans and by machines, several factors have not yet been studied. One of these factors is the effect of the transmission system (over which the message is communicated) on the accuracy with which verification is achieved. This factor is potentially an important one for digital communications problems over telephone lines where the transmission system could vary from one which gives a high-quality coded representation of the signal (e.g., log pcm) to a low-bit-rate vocoder. The purpose of this paper is to demonstrate the effects of three speech transmission systems on verification accuracy by human listeners. It is shown that the false alarm rate (i.e., a customer is rejected) is significantly higher when the test and reference utterances are transmitted by different systems than when transmitted by the same system. The miss rate (i.e., an imposter is accepted) is not significantly different for similar comparisons except for one of the conditions. The overall conclusion of this experiment is that speaker verification by human listeners cannot be performed as accurately over mixed speech transmission systems as over the same transmission system.