Interactivity-Based Quality Prediction of Conversations with Transmission Delay

While the experienced quality of a conversation transmitted over a telephone network is dependent on the parameters of the network and possible degradations, it has been shown that the sensitivity to these degradations is influenced by the type of the conversation and its contents. Especially for delayed speech transmission the conversational scenario and the adaption of turn-taking behavior of the interlocutors factor into the conversational quality rating of a specific conversation. Parametric Conversation Analysis has been proven to be a good method to extract parameters from a recorded conversation that are representable of the interactivity. While the narrowband version of a popular and standardized quality prediction model, the E-model, uses the interactivity to predict a conversational quality MOS for the given conditions, there are no attempts to predict the conversational quality of individual conversations so far. In this paper, we propose a model to predict the quality of a conversation under the influence of transmission delay based on the interactivity parameters extracted from that conversation. We evaluate which parameters are most suited for such a prediction, and compare our results to the parameter-based predictions of the E-model.