New models predicting conversational effects of telephone transmission on speech communication quality

In the planning process of telephone networks, it is important to have estimates of the speech communication quality which can be obtained with specific network configurations, even before the network has been set up. For this aim, network planning models have been developed. They base quality predictions on instrumentally measurable input parameters of the transmission chain, and take conversational factors into account. In this paper, the most prominent and complete model, the so-called E-model, is presented in more detail. The underlying principles are discussed, and it is shown how new speech processing equipment (e.g. codecs) can be included in the model. Comparisons of model quality predictions to auditory test data show how conversational aspects are covered by the model, and where modifications can improve quality prediction output. An outlook identifies the parameters where work is still needed in order to use network planning models to reliably predict conversational impacts in modern network scenarios, and to estimate the influence of the transmission channel on dialogue system quality.