The presented study follows a two fold evaluation approach in order to link NGMN characteristics to user perception. This approach is presented in Figure 1. It is based on speech samples processed under specified experimental conditions, using the Mobisense testbed. According to the twofold evaluation approach, the process is divided into two orthogonal stages - perception and networking analysis. On the perception layer, the audio samples are recorded and judged in an auditory experiment to obtain average quality scores (MOSs). On the networking layer, network traces are collected on the involved network stations and used for a subsequent network trace analysis. Finally, the results of both layers are merged to develop a quality prediction model that provides speech quality estimates, including the impact of mobility events in NGMNs. Design of such models enables a successful service adaptation and mobility management in real-time.
Figure 1: Twofold speech sample evaluation approach for NGMNs
It has to be emphasized that this experimental study is limited to the listening-only situation. The reason is that we would like to first understand what happens in the passive context, before stepping to a more interactive situation (i.e. conversational tests). As it was mentioned, interactive scenarios are more realistic, but they are also more difficult to evaluate because the results of conversational tests depend on a number of influencing factors, like the interactivity of the test scenario (free vs. more structured task-oriented conversation) and the instruction to the test subjects (i.e. whether they have been made aware of the fact that delay may occur). Thus, despite the online capability of our testbed, we left the evaluation of delay effects for further study.