Human Speaker Identification Performance Under Channel Degradations

Abstract

The effects of different transmission channel impairments on the human speaker recognition performance are assessed in this chapter by conducting two listening tests. Comparisons between NB, WB, and SWB channels are shown, as well as how the human performance is affected by the degradations introduced by speech coding, random packet loss, and electro-acoustic user interfaces in sending and in receiving direction. The participants of the listening tests listened to known voices of different lengths, distorted according to the different conditions considered. After listening to each segment, they were asked to select the speaker to whom the heard stimulus corresponded from a set of given options. Closed-set speaker identification of familiar voices was studied, where the true speaker is known to be in the target speaker group. This is a more realistic situation for telephone service users than initiating a conversation with a previously unknown person. The listeners and the speakers asked to participate in the tests were work colleagues at the same department and knew each another from their long-term acquaintance. The voices of the speakers were collected through a high-quality microphone in clean conditions and subsequently transmitted through channels with the different degradations to be examined in the tests.